research and write an article 2 pages long on the Applications of Random Forests Algorithm on Illicit Supply Chain Network.

Unconfirmed95964.crdownload2021-IISET-CFP-DisruptingIllicitOperations004 s13388-017-0029-81
 

research and write an article 2 pages long on the Applications of Random Forests Algorithm on Illicit Supply Chain Network. 

Full Terms & Conditions of access and use can be found at
https://www.tandfonline.com/action/journalInformation?journalCode=uiie21

IISE Transactions

ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/uiie21

A review of research in illicit supply-chain
networks and new directions to thwart them

Rashid Anzoom, Rakesh Nagi & Chrysafis Vogiatzis

To cite this article: Rashid Anzoom, Rakesh Nagi & Chrysafis Vogiatzis (2021): A review of
research in illicit supply-chain networks and new directions to thwart them, IISE Transactions, DOI:
10.1080/24725854.2021.1939466

To link to this article: https://doi.org/10.1080/24725854.2021.1939466

Published online: 06 Aug 2021.

Submit your article to this journal

Article views: 346

View related articles

View Crossmark data

https://www.tandfonline.com/action/journalInformation?journalCode=uiie21

https://www.tandfonline.com/loi/uiie21

https://www.tandfonline.com/action/showCitFormats?doi=10.1080/24725854.2021.1939466

https://doi.org/10.1080/24725854.2021.1939466

https://www.tandfonline.com/action/authorSubmission?journalCode=uiie21&show=instructions

https://www.tandfonline.com/action/authorSubmission?journalCode=uiie21&show=instructions

https://www.tandfonline.com/doi/mlt/10.1080/24725854.2021.1939466

https://www.tandfonline.com/doi/mlt/10.1080/24725854.2021.1939466

http://crossmark.crossref.org/dialog/?doi=10.1080/24725854.2021.1939466&domain=pdf&date_stamp=2021-08-06

http://crossmark.crossref.org/dialog/?doi=10.1080/24725854.2021.1939466&domain=pdf&date_stamp=2021-08-06

A review of research in illicit supply-chain networks and new directions to
thwart them

Rashid Anzoom , Rakesh Nagi , and Chrysafis Vogiatzis

Department of Industrial and Enterprise Systems Engineering, University of Illinois at Urbana-Champaign, Urbana, IL, USA

ABSTRACT
Illicit trades have emerged as a significant problem to almost every government across the world.
Their gradual expansion and diversification throughout the years suggests the existence of robust
yet obscure supply chains as well as the inadequacy of current approaches to understand and
disrupt them. In response, researchers have been trying hard to identify strategies that would suc-
ceed in controlling the proliferation of these trades. With the same motivation, this article con-
ducts a comprehensive review of prior research in the field of illicit supply-chain networks. The
review is primarily focused on the trade of physical products, ignoring virtual products and serv-
ices. Our discussion includes analyses of their structure and operations, as well as procedures for
their detection and disruption, especially from the perspective of operations research, manage-
ment science, network science, and industrial engineering. We also address persisting challenges
in this domain and offer future research directions to pursue.

ARTICLE HISTORY
Received 27 November 2020
Accepted 26 May 2021

KEYWORDS
Illicit trade; supply-chains;
disruption of illicit
networks; literature review;
future directions

1. Introduction

Over the last few decades, the world has experienced unprece-
dented growth in commerce, spanning across different coun-
tries and continents. This growth has outpaced the existing
governance mechanisms, resulting in the proliferation of illicit
trades. Despite the adoption of numerous measures, govern-
ment entities have fallen short of halting the growth of such
trades, which now make up approximately 8–15% of the glo-
bal GDP (Mashiri and Sebele-Mpofu, 2015). This calls for a
better understanding of illicit trade and its operations. To aid
in this ensuing battle, researchers from different disciplines
have come forward to contribute to this field. Aligned with
this perspective, we are presenting a literature review on the
operation and disruption of illicit trade, which we believe will
prove useful to the policymakers and fellow researchers in the
field of operations research, management science, and indus-
trial engineering.

The field of illicit trade is a vast one that can be catego-
rized by product, market, or trade characteristics. Existing
works have mostly focused on a particular category of illicit
trade, e.g., literature review of counterfeit trade by Staake
et al. (2009). Others have focused on a specific aspect of the
trade. Bichler et al. (2017) reviewed the literature related to
the network structure of drug trafficking organizations.
Kammer-Kerwick et al. (2018) outlined the application of
operation research and data science in combating human
trafficking. In contrast, we intend to discuss the domain of
illicit trade in a holistic manner, comprising both qualitative
and quantitative aspects. It is done in two ways. First, we try
to picture the operations of illicit trade from two perspectives:

supply chain and network analysis. Second, we present meth-
odologies, especially in the field of operations research and
data science that have been proposed to help combat the pro-
liferation of these trades. We also look into the research gaps
and suggest future research directions to pursue.

The organization of this article is as follows. We start
with a presentation of the selection criteria of the literature
and general statistics in Section 2. Section 3 provides a big
picture discussion on illicit trades. Sections 4 and 5 review
illicit activities from the supply chain and the network per-
spective, respectively. Section 6 is devoted to different meth-
odologies used to identify entities related to illicit trade.
Section 7 discusses strategies to combat illicit activities and
their associated networks. Section 8 is a critical analysis into
the research gaps and possible directions for future research.
Finally, Section 9 concludes the discussion with summary
statements. Following this sequence is not mandatory; in
fact, one can move from Section 3 to any of the other sec-
tions based on your interests. For example, a reader more
interested in the network perspective and less in supply
chain aspects can skip Section 4 and directly proceed to
Section 5.

2. Review methods and statistics

To our best knowledge, there has not been any review paper
discussing illicit trades/supply chains on such a broad scale.
The topics addressed in the review comprise research from
multiple disciplines (industrial engineering, management sci-
ence, criminology, network science). As a result, we could

CONTACT Rakesh Nagi nagi@illinois.edu
Copyright � 2021 “IISE”

IISE TRANSACTIONS
https://doi.org/10.1080/24725854.2021.1939466

http://crossmark.crossref.org/dialog/?doi=10.1080/24725854.2021.1939466&domain=pdf&date_stamp=2021-08-05

http://orcid.org/0000-0003-4699-6379

http://orcid.org/0000-0003-4022-6277

http://orcid.org/0000-0003-0787-9380

https://doi.org/10.1080/24725854.2021.1939466

http://www.tandfonline.com

not set any specific strategy for searching the papers.
Instead, we had to rely on searching through Google Scholar
for a set of keywords related to illicit trades (e.g., illicit/
illegal trade, illicit/illegal supply chain/network). We also
searched for keywords relevant to individual trade categories
(drugs, counterfeit, arms, wildlife). The initial screening was
done through reading the abstract. While reading the
selected manuscripts in detail, additional papers were dis-
carded or included in the literature as per relevance. The
final tally of the articles cited in this article stands at 239
and their sources include journals, conference proceedings,
online and technical reports, and books. A summary of the
descriptors of the literature reviewed is shown in Figure 1
(Year-wise statistics), Figure 2 (Trade-wise statistics), and
Figure 3 (Journal statistics).

As seen in Figure 1, research in illicit trades has increased
gradually over time, with approximately a 40% increase over
the last 5 years. In terms of trade categories, the top position
is reserved, as expected, for narcotics. However, the glaring
gap between narcotics and the other trades shows the skew-
ness of research advancement and focus in illicit supply
chains. Figure 3, on the other hand, indicates the highly
diverse perspectives of illicit supply chain analysis, including
criminology, operations research, data science, network sci-
ence, risk analysis, and so on. The highest number of cita-
tions from a single journal (Crime, Law, and Social Change)
was only six, which is only 2.5% of the total number
of articles.

3. Illicit trade

3.1. Definition of illicit trade

According to the World Economic Forum (2012), illicit
trade involves the process of gaining money, goods, or value
gained from illegal and generally unethical activity causing
harm to the economy, society, environment, or politics.
Feige (1997) described illicit trades comprising non-compli-
ant economic behaviors like evasion, avoidance, circumven-
tion, and corruption of rules as well as efforts to hide these
behaviors from public authority surveillance. On the

contrary, governmental institutions try to suppress these
trades leading to social, economic, and organizational fric-
tion between the two entities. Crotty and Bouch�e (2018)
mentioned two key risks distinguishing illicit markets from
its licit counterparts. First, buyers and sellers carry the risk
of getting arrested. Second, they cannot rely on state or legal
institutions to enforce market rules. These risks prevent
illicit businesses from adapting to changes in the economic
environment, rendering their failure to match the efficiency
of licit trades (Dean et al., 2010). Despite such inefficiency,
illicit trade is still maintaining an annual turnover of 2.2 tril-
lion dollars (Coke-Hamilton and Hardy, 2019).

3.2. Reasons behind illicit trade

While investigating the factors facilitating illicit trade, one
must first acknowledge that significant demand exists for its
associated products. This demand, along with the lucrative
payoff for successful transactions, creates strong economic
incentives for participation in illicit trades (Basu, 2014a).
Poor socio-economic condition adds stability to the market,
as people start treating this as a profession. And the growth
or decline of the market is dictated by risk in operation,
which in turn is dependent upon the government’s ability
(or will) to detect and prosecute criminals (Helbling et al.,
2012; Grant Thornton, 2013; Hauenstein et al., 2019). Apart
from these, regional influence, trade regulation, tax policy,
and lack of awareness also contribute to its growth
(Helbling et al., 2012; Basu, 2014a; Patel et al., 2015).
However, one should not consider this list as exhaustive
since there can be factors specific to a particular trade cat-
egory or a country. Statistical hypothesis testing could be
one way to identify these additional factors. Gonz�alez
Ordiano et al. (2020a) recently proposed another approach
using node embedding and clustering.

Researchers have provided several quantitative models
regarding the growth of the illicit market. Caulkins and
Padman (1993) pictured the narcotics market growth as a
function of the associated utility and risk, whereas Baveja

Figure 1. Summary of research on illicit supply chain over time.
Figure 2. Summary of research on illicit supply chain by trade.

2 R. ANZOOM ET AL.

et al. (2004) based it on the enforcement level and economic
hardship. Koen et al. (2017) attempted to predict wildlife
trafficking through causal modeling, but had to rely on
expert judgment for the modeling, due to data scarcity. The
Economist Intelligence Unit Limited (2018) developed an
index to indicate the vulnerability of a country to illicit
trade. The index, however, was linear in nature and did not
consider possible interrelationships between the factors.

3.3. Classification of illicit trades

Classification of illicit trades is roughly based on two prem-
ises: product and trade characteristics. Researchers and gov-
ernment agencies have mostly adopted the former approach,
although their categorization has not been uniform (Basu,
2014a; WCO, 2017; Transnational Alliance to Combat Illicit
Trade, 2019). The products are usually considered as either
physical or virtual. However, illicit trades also involve the

egregious act of human and wildlife trafficking, which does
not go with the definition of products and thus require add-
itional categories. Illegality may also arise from a particular
aspect of the trade: the product, its acquisition, exchange, or
other regulation breaches (Beckert and Wehinger, 2013).
These characteristics form the basis of an alternate classifica-
tion approach. Fuzing both approaches, Staake et al. (2009)
outlined a comprehensive classification scheme for illicit
products and services. The current article adopts a similar
approach, but overlooks services and virtual products.
This intensifies our focus to the trade of physical goods,
which is further classified into four categories: contraband
trade, counterfeit trade, fencing, and parallel/illicit import.
Figure 4 denotes the classification scheme.

The first two categories to appear in our classification are
the trade of contrabands and counterfeits. Contrabands are
products with an embargo or restriction on production (e.g.,
narcotics) or/and distribution (e.g., arms). Counterfeits, on

Figure 3. List of journals with multiple publication on illicit trades.

IISE TRANSACTIONS 3

the other hand, mimic the characteristics of a brand prod-
uct. They can be of two types: deceptive and non-deceptive
(Staake et al., 2009; Cho et al., 2015). Non-deceptive coun-
terfeits can be distinguished from the brand products and
are sold at a discount. In contrast, deceptive counterfeits are
hard to detect and sold at the same price as brand products.
The third category, fencing, represents the trade of stolen
products (Johns and Hayes, 2003). The product traded can
be used or new, and occasionally with an alteration (e.g., car
parts). The final category, parallel import, is the sourcing of
a legal product without authorization of the intellectual
property owner. Products subjected to excise tax (e.g., cigar-
ette, alcohol) commonly dominate this trade (opportunity to
evade tax by importing from lower tax region). The legality
of these products, however, is a matter of controversy and
often hinges upon the law of exhaustion of intellectual prop-
erty rights (Williams, 2020). For simplicity, we consider all
such trades as illegal. It might also be possible for the same
product to be distributed through different types of trades.
For example, tobacco products in the market can be coun-
terfeits, illegally imported, or even fenced. Moreover, trade
can also occur physically as well as virtually. All these varia-
tions make the general analysis of illicit trades somewhat
challenging.

3.4. Impact of illicit trades

The impact of illicit trades is multi-faceted, traversing across
different viewpoints. Illicit trade undermines human rights,
upsets ecological balance, and circumvents law and order of
the country. Another possible consequence is the loss of
consumer welfare. Some of the products traded (e.g., nar-
cotics, counterfeit drugs) can directly harm one’s well-being.
The lack of commitment to maintain quality or provide ser-
vice raises the risk of mismatch between actual and expected
utility. Proactive consumers who assume this risk may opt
not to buy such products or expect a lower utility from the
product (Cho et al., 2015). These hurt the profitability of
legal organizations in terms of lost sales, decreased brand
value, and increased R&D expenditure. As a result, the gov-
ernment collects less tax revenue. This, coupled with the
poor performance of licit organizations, can cause
unemployment creating further incentives for participation
in illicit trades. For further information, we refer to the

work by Hintsa and Mohanty (2014) and Transnational
Alliance to Combat Illicit Trade (2019), which discuss the
implications of illicit trades from socio-economic and global
sustainable development perspectives.

3.5. Sources of data on illicit trade

Acquisition and analysis of data is useful for developing a
realistic understanding of illicit operations and their associ-
ated network. However, significant challenges need to be
overcome during their collection and application (e.g., avail-
ability, accuracy, completeness), hindering development of
quantitative studies. Researchers have nevertheless attempted
to produce them using both conventional and innovative
sources. For convenience of the readers, some of these data-
bases are listed in Table 1.

Based on accessibility, one can categorize data sources as
open, closed, and classified. Data provided by open sources
are publicly accessible, whereas those in closed sources require
permission for access. Classified sources contain vital informa-
tion that are restricted from sharing. From a format perspec-
tive, data provided by these sources can be classified as
unstructured, semi-structured, and structured. Unstructured
data contain a swarm of information, whereas structured data
systematically categorize the information. Organization in
semi-structured data falls somewhere between these two. The
following subsections introduce five major sources of data and
discuss their utility and limitations. These include: news and
media, law enforcement, court proceeding, organizational
database, and miscellaneous sources.

News and media: Whenever a significant event such as seiz-
ure or arrest occurs, it appears in the media. They also pro-
vide investigative reports occasionally. Coscia and Rios
(2012) used such reports from online newspapers and blogs
to identify the mobility of drug traffickers. Important infor-
mation can also be obtained from different websites (Crotty,
2015; Patel et al., 2015; Farrugia et al., 2020). In recent
times, researchers have turned to social media for data accu-
mulation (Mackey and Kalyanam, 2017; Zhao et al., 2020).
Although easy access is a definite advantage for these sour-
ces, most of the data available are unstructured and require
further processing. Moreover, the reliability of these data is
questionable.

Figure 4. Classification of illicit trade.

4 R. ANZOOM ET AL.

Law enforcement: Due to direct engagement with criminals,
law enforcement agencies possess substantial information
regarding illicit networks. These information are mostly
unstructured and available in multiple forms including arrest
records (Morselli and Petit, 2007; Duijn et al., 2014), phone
records (Agreste et al., 2016), wiretap transcripts (Natarajan,
2006), intelligence reports (Malm and Bichler, 2011; Coutinho
et al., 2020; Toledo et al., 2020), and co-offense and financial
transaction reports (Levitt and Venkatesh, 2000). However,
they are not readily attainable and classified in some cases, and
often depend on the extent of the investigation and might
include bias. Nevertheless, this remains a major data source for
researchers.

Court proceedings: Researchers have also used court proceed-
ings as a source of data for illicit trade, often in the format of
the prosecutor’s file and court records (Fuentes, 1998; Becucci,
2004). These involve the summary of law enforcement investi-
gation (Agreste et al., 2016; Cavallaro et al., 2020), in addition
to witness statements, transcripts of trial, and judges’ sentenc-
ing comments (Bright and Delaney, 2013; Bright et al., 2019).
The advantage of using this source is that it contains several
types of information. And since they are made available after
closure of the case, access is relatively easy. However, the time
difference between the event occurrence and data availability
can be long, depending upon the pace of prosecution. And
similar to law enforcement data, sampling bias is possible,
since available data mostly represent information of failed
(busted) enterprises, not necessarily the successful ones.

Organizational databases: Various organizations are work-
ing nationally and internationally on different aspects of
illicit trades (e.g., OECD, WCO, EMCDDA, UNCTAD,
UNODC, WHO). Many of these maintain databases that are
mostly structured and closed. Researchers have often used
these data in their research (e.g., UNODC Individual Seizure
Data by Giommoni et al. (2017), Consolidated Counterdrug
Database by Magliocca et al. (2019)). Business organizations
can also keep databases to track their products and detect
possible counterfeits. Gonz�alez Ordiano et al. (2020b) used
such a dataset for analyzing licit and illicit supply chains.

Miscellaneous: This includes any sources that do not fall
under the above categories. An interesting instance was
Tsirogiannis and Tsirogiannis (2016) using the book of
Watson and Todeschini (2007) to derive illicit antiquity traf-
ficking networks. Personal interviews are also often used for

information extraction (Stevenson and Forsythe, 1998;
Bradshaw, 2016; Caulkins et al., 2016;).

As previously mentioned, data on illicit trade is far from
being perfect. Existing flaws relate to four major issues:
incompleteness, boundary specification, dynamics, and coord-
ination. The first one, incompleteness, is not surprising, given
the concealment of illicit networks. Without complete infor-
mation, analysis of these data may suffer from lower credibil-
ity. Researchers, however, have made progress in predicting
some of the missing entities (see Section 6.2). The second
issue, also known as the boundary specification problem,
addresses the confusion regarding the extent of the network
to consider. In response to this question, Bouchard (2007)
presented three viewpoints: research intent, member outlook,
and social interactions. Campana and Varese (in press) coun-
tered by suggesting five strategies regarding boundary specifi-
cation. However, the information available is collected from
the perspective of law enforcement agencies and thus may not
meet these conditions, leading to possible bias. The third con-
cern is about the incorporation of dynamics, which requires
frequent updating of the database. Generally only organiza-
tional databases happen to do so, since other sources provide
event-specific data. Understandably delayed access may make
the data obsolete, diminishing the value of its insights. The
last concern we raise is the coordination of data among differ-
ent organizations. Nowadays, multiple agencies gather data on
distinct features of illicit trades. Often there is an overlap in
the trade or jurisdiction. Data sharing would accelerate this
data accumulation process. However, agreements must be
made on terminologies to use. Haas and Ferreira (2015) illus-
trated the development of such a database on wildlife traffick-
ing. Another possible coordination approach could be the
amalgamation of different types of data (Kammer-Kerwick
et al., 2018).

4. Supply chain view of illicit trade

A typical supply chain consists of five stages: supplier,
manufacturer, distributor, retailer, and consumer (Chopra
and Meindl, 2019). For illicit supply chains, though, no such
exact number has been universally agreed upon. Kilmer and
Hoorens (2010) proposed the existence of four stages (pro-
duction, distribution, retail, and consumers) for narcotics
supply chain, whereas Basu (2014b) listed six for wildlife
trafficking. The latter further split the stages into three
phases: upstream activities, concealment, and distribution.
Most of the activities described in these stages were congru-
ent with a conventional supply chain model. This led us to

Table 1. Databases related to illicit trade.

Database Context Author/Year Accessibility

MAGLOCLEN Survey Drug Caulkins (1995) Closed
Medicine Quality Database Countereit Isah et al. (2015) Closed
HealthMap Wildlife Trade Wildlife Patel et al. (2015) Open
Drug Retail Price Data Drug Caulkins et al. (2016) Open
UNODC Individual Drug Seizure Cases Drug Giommoni et al. (2017) Closed
Consolidated Counterdrug Database Drug Magliocca et al. (2019) Closed
Network Disruption Data Mafia Cavallaro et al. (2020) Open
Link of Occurrence Database Criminal Toledo et al. (2020) Closed
Etharscamdb Crypto-currency Farrugia et al. (2020) Open

IISE TRANSACTIONS 5

adopt the aforementioned five stages in the illicit supply
chain model and we added a new stage on smuggling. A
fundamental requirement in the illicit supply chain is the
veiled flow of products and services, which is satisfied by
smuggling. Turner and Kelly (2009) defined smuggling as
the clandestine and unlawful transportation of goods
between different jurisdictions. Viewed simplistically, its
function is the same as logistics, i.e., transferring products
from one location to another. However, it also involves tasks
(e.g., product concealment, evasive route selection, docu-
ment forgery, corruption of officials) deemed indispensable
for success. Thus, smuggling can be considered a core com-
petence in the illicit supply chain (Basu, 2013). The pursuit
of excellence in this domain has eventually contributed to
the emergence of smugglers, i.e., independent professionals
specializing in smuggling (Morselli, 2001; Basu, 2013).
Acknowledging these unique attributes, we consider smug-
gling as a distinct stage in the illicit supply chain.

As mentioned in Section 3.1, not all stages in the supply
chain need to be illegal. Here we define a step to be illicit if
its actors participate consciously in the supply chain. Based
on this definition, legal status of the stages might vary across
trades. For example, suppliers are illicit in fencing, but licit
in parallel trade (since the product manufactured is legal).
One or more stages might be missing for some trades as
well (e.g., unaltered fencing). Such disparity generates differ-
ent flow configurations within the supply chain, as shown in
Figure 5. Finally, variation may also stem from the direction
of flows. The usual norm is the forward flow of products

and the reverse flow of funds. However, some cases might
involve bi-directional flow of products (e.g., exchange of
products in fencing), or an entirely different network for
fund flow (Johns and Hayes, 2003; Brown and Hermann,
2020). Researchers have tried to incorporate these differen-
ces through multiple models, which we discuss in the fol-
lowing section.

4.1. Modeling perspectives of the illicit supply chain

Existing modeling approaches for the illicit supply chain
have been conducted from the perspective of a particular
trade rather than in a comprehensive manner. In this regard,
the field of narcotics has received by far the most attention.
Caulkins (1997) developed a model to describe the domestic
distribution network of narcotics. It represented the number
of customers to serve (branching factor) as a function of the
quantity discount factor for price markup and the ratio of
selling costs to product costs. Other modeling approaches of
narcotic supply chains include network flow representation
by Helbling et al. (2012), cyclic view by Caulkins et al.
(2016), and global production network framework (Dicken,
2003) by Miltenburg (2018). Markowski et al. (2009) devel-
oped a probabilistic multi-channel supply chain model to
demonstrate the trade of illicit small arms. To explain the
robustness of the supply chain, they presented the concept
of tie and cut set, which denoted the minimum number of
elements required to connect and disconnect the supply
chain, respectively. For analyzing the trafficking of nuclear

Figure 5. Supply chain flow in different illicit trades.

6 R. ANZOOM ET AL.

products, Bradshaw (2016) proposed Illicit Non-state
Nuclear and Radiological Trafficking network model. The
model incorporated features of three existing models: loca-
tional model, network model, and enterprise model. The
locational model looks into the factors that facilitate the ori-
gination of a network, the network model visualizes the flow
within the network, and the enterprise model outlines four
major components governing the flow: supply, regulators,
competition, and customer. Apart from these, Stevenson and
Forsythe (1998) discussed four disposal methods of stolen
goods (supply chain configuration) in fencing. Recently,
Gonz�alez Ordiano et al. (2020b) used a variable-state reso-
lution Markov Chain to model both licit and illicit (counter-
feit) supply chains. Here they identified the optimal set of
states (geographic areas) across three levels of analysis
(country, region, continent) that best fit the supply chain
model. Comparison of the limiting distributions for licit and
illicit supply chain led to the discovery of potential hotspots
of counterfeit activities.

4.2. Associated revenue and cost

As discussed in the previous section, illicit trade is quite
profitable, with the price markup of some products reaching
600%. However, the distribution of revenues across different
stages is not properly demarcated. A substantial share of the
revenue is thought to be enjoyed by the distributors, grant-
ing them power over manufacturers and retailers (Caulkins
et al., 2016; Miltenburg, 2018).

Caulkins and Padman (1993) discussed five areas of
expenditure in narcotic supply chains: procurement, trans-
portation, inventory (holding and stock-out), and risk of
arrest. Clemons et al. (1993) categorized them into two
components: coordination cost and cost due to transaction
risk. The first one is incurred in activities between trade
partners. And the second one is due to risks regarding
detection of operations, seizure of products and equipment,
and arrest of members. To mitigate these risks, investment
is made in concealment, corruption, and evasion. Basu
(2014b) adopted these as the three elements of transaction
cost. Williamson (1975) listed four drivers influencing these
costs: asset specificity, operational uncertainty, frequency of
operation and interdiction, and human behavioral aspects.
Significant cost also occurs in money laundering, i.e., 10 to
17% of the smuggled amount (Soudijn and Reuter, 2016).

4.3. Performance drivers and strategies

Like legal business organizations, stakeholders in illicit trade
also aim for efficiency to increase profit. However, pressure
from law enforcement agencies forces them to incorporate
strategies to secure their supply chain operations, creating a
tradeoff between the two objectives. Chopra and Meindl
(2019) listed six drivers of traditional supply chain perform-
ance: facility, inventory, transportation, sourcing, pricing,
and information. Basu (2014b) introduced three additional
features specific to an illicit supply chain: transnational
smuggling, money laundering, and corruption. We, however,

consider smuggling as a stage rather than a performance
driver. Along with the remaining eight, we introduce two
new drivers: concealment, and innovation and technology.
In the following sections, we discuss the strategies regarding
these 10 drivers.

1. Facility: Facilities in illicit trades are discreet and
mobile, creating difficulty in their detection (Basu,
2014a). Despite this limitation, researchers have
attempted to investigate their location strategies.
Stevenson and Forsythe (1998) identified different stor-
age facilities used in fencing. Zhao (2019) analyzed the
facilities involved in the supply chain of chemicals that
are precursors to narcotics in China. They noted the
geographic location of manufacturers to be concen-
trated in border provinces and rural settings, which
indicate the consideration of efficiency and security,
respectively. Crotty and Bouch�e (2018) went ahead
and studied the locational strategies of illicit massage
parlors linked with sex trafficking. The authors used
Gi� clusters (Mitchel, 2005) to identify the spatial clus-
ters of these facilities, whereas ordinary least squares
regression and geographically weighted regression were
applied to determine the variables impacting their
configurations.

2. Inventory: Similar to legal trades, illicit trades also
face the risk of a mismatch between demand and sup-
ply, leading to the necessity of inventory management.
Inventory is also needed to “cool off” stolen products
to avoid detection by law enforcement agencies (Johns
and Hayes, 2003). Not much work is found concerning
the inventory management issues of illicit supply
chains. Basu (2013) stated different responsibilities in
narcotics inventory management that are carried out
by stash managers. Caulkins et al. (2016) observed an
increase in inventory turnover ratio down the drug
supply chain. Magliocca et al. (2019), on the other
hand, found the lot-size to be inversely related to
interdiction risk. Miltenburg (2018) calculated an
inventory level equal to 2 years of productions for opi-
ate and heroin supply chains. However, his empirical
study found a lower inventory level in practice, along
with the absence of safety stock. These combined with
seizure by law enforcement and variation in annual
production can exacerbate the risk of shortage. For
some narcotics such as heroin, this shortage is miti-
gated by reducing the purity of the product.

3. Transportation: Success in illicit trade depends heavily
on the execution of its logistics function, especially in
concealment and evasion. While selecting an appropri-
ate strategy, criminals have to make decisions regard-
ing the modes and routes of transportation, as well as
the number of transshipment points. Two factors gov-
ern these decisions: profit and risk. For example, limit-
ing the number of transshipment points allows
members to enjoy a higher profit share. However, it
also increases the distance of a shipment leg, and
hence, the risk of interdiction. Being well versed in

IISE TRANSACTIONS 7

these affairs, illicit traders adopt strategies befitting dif-
ferent scenarios (Basu, 2014a).

Researchers have characterized the distribution of
smuggling routes to be non-uniform and clustered
across a small number of paths (Boivin, 2013;
Arroyave et al., 2020). Using exponential random
graph modeling, Giommoni et al. (2017) identified
four factors behind such configurations: risk, profit,
geographic and social proximity. The presence of
adversaries (e.g., rival organizations, law enforcement
agencies) can also play a role in shaping the routes.
Several attempts have been made to model the routing
decision of smugglers. Sin and Boyd (2016) applied
Dijkstra’s algorithm (Dijkstra, 1959) to define the
potential route of nuclear traffickers that maximized
their probability of evasion: specifically, they used a
proxy for probability of successfully traversing an edge
as its corruptibility. Meneghini et al. (2020) created a
geographical likelihood network (where edges denoted
the likelihood of flow between two node) based on
seizure and consumption data of illicit cigarettes. The
best set of paths in this network was identified using
Yen’s algorithm (Yen, 1970). Tsirogiannis and
Tsirogiannis (2016) tested three additional algorithms
to infer the smuggling routes of antiquities between
two given locations: shortest path algorithm, local
expansion algorithm, and maximum weight arbor-
scence method. The characteristic of the optimal path
in these three methods were: minimum number of
links, presence of most probable links, and maximin
transaction volume, respectively. Two criteria were
used to evaluate their accuracy: number of correct pre-
dictions and modified Levenshtein distance (Wagner
and Fischer, 1974), i.e., the minimum number of node
operations required in the predicted path to match the
correct one. Shortest path algorithms appeared to be
the better performer. Magliocca et al. (2019) developed
an agent-based model to simulate the shift of routes in
response to government interdiction.

4. Pricing: Most of the studies addressing illicit product
pricing take place in the field of narcotics. One of the
earliest works regarding this was done by Brown and
Silverman (1980), which estimated the retail price of
heroin based on quantity and purity. Caulkins (1995)
found drug prices to be an increasing function of the
distance between sales point and source, as well as a
decreasing function of market size and transaction
quantity. However, their findings were limited to local
markets only. Miltenburg (2018) dismissed this reser-
vation by stating that the retail price was determined
solely by local factors. Despite these formulations, drug
prices exhibit high volatility, due to low inventory lev-
els and uncertain lead times (Caulkins and Baker,
2010).

Cho et al. (2015) presented a game-theoretic model
to simulate the pricing decision of counterfeit products
in retail and wholesale stages. Here, equilibrium strat-
egies (subgame-perfect Nash) were derived for

counterfeiters in response to the quality and price of
brand products. Other factors taken into consideration
for determining the optimal price involved functional
quality and market share of the counterfeit product, as
well as the risk and cost of product confiscation or
dealer incarceration.

5. Sourcing: Sourcing of precursor materials plays a piv-
otal role in ensuring the viability of the illicit supply
chain. These materials may be legal or illegal. Legal
ones are known as dual goods, since they have applica-
tion in both businesses. Criminals often use deceptive
tactics to procure these from suppliers without expos-
ing their intended use (Albright et al., 2010; Zhao,
2019). However, suppliers can also conspire with the
criminals. The number of suppliers to include may
vary depending on the sourcing strategy. However,
there exist risks of incompetence and betrayal since no
legal binding is applicable. Johns and Hayes (2003)
discussed this issue in fencing and categorized the sup-
pliers into four groups based on their competence and
reliability: good thieves, in-between thieves, walk-in
thieves, and dopers. Liu et al. (2004), on the other
hand, modeled the sourcing decision of a distributor
as a shocked multi-item newsvendor problem, where
both genuine and counterfeit products could be pur-
chased. In this model, optimal purchase quantity was
determined by inspection rate and penalty.

6. Information: The performance of a supply chain is
highly dependent on its communication structure.
Better information sharing can improve coordination
across different stages. The same goes for the illicit
supply chain. However, the constant threat of interdic-
tion makes the dissemination of information risky
here. Illicit traders, hence, face a tradeoff between effi-
ciency and security while determining the extent of
information sharing. One way to accomplish this is to
compartmentalize operations, restricting the spread of
its information only to specific groups. This way, if
any member is exposed, only information about that
particular operation will be imperiled.

Regular success in extracting information from
phone record and wiretap denote that criminals use
conventional technologies like cellphones. However,
the use of advanced communication technologies is
also increasing (Deville, 2013). Over the last few years,
the internet has appeared as an essential communica-
tion medium. Criminals are observed using chat rooms
and private messaging services to exchange informa-
tion about illicit operations (Motoyama et al., 2011).
Social media and darknet are also facilitating commu-
nication between vendors and consumers.

7. Concealment: Criminals employ different concealment
strategies to avoid detection by adversaries (e.g., law
enforcement agencies, licit suppliers and retailers). The
bulk of their efforts focus on the clandestine transpor-
tation of illicit products. This involves embedding
them into licit products, hiding in vehicles, or using
misleading packaging (Williams and Godson, 2002;

8 R. ANZOOM ET AL.

Decker and Chapman, 2008; Zhao, 2019). Basu (2013)
documented different concealment strategies used in
drug smuggling. The adoption of a particular conceal-
ment strategy depends on its implementation cost as
well as the cost inflicted to government agencies for
their detection. If the second cost exceeds the
budget allocated for law enforcement, the detection
risk will reduce for the illicit supply chain, increasing
its profitability in turn. For these reasons, concealment
capability can be considered a key driver in the illicit
supply chain.

8. Corruption: The aim of corruption is to dissuade law
enforcement agencies from disrupting illicit trade
(Basu, 2014b; Shelley, 2018). In exchange, the corrupt
personnel receives periodic bribes or a cut from unit
revenue. Criminals see this as a competitive advantage,
as it allows them to use cost-efficient transportation
routes, enjoy lenient or no inspection, and protect ter-
ritorial integrity (Michael, 2012). In addition, corrupt
agents can also act as double agents providing infor-
mation about law enforcement plans (Van Der Veen,
2003). This relationship between criminals and officials
can be short-term and activity-specific, or long-term
and institutionalized. Criminals favor the latter since it
allows them greater control over operations. And at its
extreme state, criminals can even influence national
policies toward their advantage (Greenhill, 2009).
These advantages, however, are weighed against costs
affiliated with corruption. Basu (2014b) discusses these
costs as well as their drivers.

9. Innovation and Technology: Every time law enforce-
ment agencies catch a shipment of illicit products, its
trafficking strategy is compromised. Criminals, there-
fore, have to conjure a new strategy for sustaining
their operations. Thus, innovation plays a crucial role
in strengthening the illicit supply chain, especially in
the concealment of goods and finances (Basu, 2013).
The rapid pace of technological advancement has
remarkably enhanced these capabilities, e.g., sales and
marketing by social media and darknet, money laun-
dering through crypto-currencies (Seddon, 2014;
Dittus et al., 2018). However, these amenities also pro-
vide law enforcement agencies information to trace
their activities. Mackey and Kalyanam (2017) detected
illicit fentanyl sale sites from twitter using text filtering
and biterm topic model. For similar purpose, Zhao
et al. (2020) used methods based on a support vector
machine and convolutional neural networks. Di Minin
et al. (2018) suggested using deep learning algorithms
to detect illegal wildlife trades. All these approaches
were implemented in contraband trade, where the
products are unique. However, in case of other illicit
trades, the classifier may face difficulty in differentiat-
ing between legal and illegal products because of high
similarity (see, e.g., the study by Sam et al. (2007) to
identify pornographic websites).

10. Money Laundering: Finance is considered a critical
resource in an illicit supply chain and thus holds the

key to organizational viability. Understanding this, law
agencies try to intercept its flow in illicit supply chains
(Bright et al., 2017). Close surveillance by officials
deters the use of traditional transaction channels for
this purpose. However, criminals circulate funds either
by skirting security measures in conventional mediums
or through innovative mechanisms (Godspower-
Akpomiemie and Ojah, 2019; Brown and Hermann,
2020). This illegal flow of money, i.e., money launder-
ing, consists of three stages: disassociation from source,
obfuscation of money trail, and legitimization of the
fund (Naylor, 2004). Brown and Hermann (2020)
denoted three primary methods of money laundering:
banking system, non-banking system (e.g., hawala,
cash courier), and geographically oriented system (e.g.,
offshore company, free trade zone).

Although non-cash payments are not uncommon, cash
remains the preferred payment method in illicit trade, due
to low traceability (Tammaro, 2014; Zhao, 2019). Of late,
cryptocurrencies are steadily gaining popularity with the
provision of anonymity and swift convertibility, greatly
enhancing the security of an illicit supply chain (Fanusie
and Robinson, 2018). In response, researchers have invested
in methods to detect and thwart money laundering. Chen
et al. (2018) provided a detailed review of machine learning
methods used to detect suspicious bank transactions. On the
other hand, Hirshman et al. (2013) and Farrugia et al.
(2020) attempted to identify accounts conducting illegal
transactions of bitcoins and etherium. To do so, the former
used k-means clustering and RolX algorithm, and the latter
employed the XGBoost algorithm.

5. Network view of illicit trade

The frequent association of illicit trade enablers with organ-
ized criminal groups and terrorists has created the need for
viewing and interpreting illicit trade operations in terms of
the interactions between the entities involved. This, in turn,
has led to a network representation of illicit supply chains,
commonly referred to as illicit or dark networks (Duijn
et al., 2014). Researchers have leveraged different tools from
network science and analysis for analyzing the properties of
these networks. In particular, social network analysis
(Wasserman and Faust, 1994) is commonly employed in the
literature on illicit networks (see, e.g., Kinsella (2008); Bright
et al. (2012); Arroyave et al. (2020)). Ideally, one would like
to know the full network in advance. However, this informa-
tion is often unavailable at the outset and has to be obtained
through a discovery process: these mechanisms are discussed
later in Section 6. For the remainder of this section, we
assume complete network knowledge, i.e., consider the com-
pletion of discovery process. In the following subsections,
we explore different features of illicit networks and present
studies relevant to them.

IISE TRANSACTIONS 9

5.1. Structure of illicit networks

Based on their structure, one may classify networks as ran-
dom Erd}os-R�enyi (ER) (Erd}os and R�enyi, 1959), small-world
(Watts and Strogatz, 1998), or scale-free (Barab�asi and
Albert, 1999). Random ER networks possess links with the
same probability of existence. Small-world networks are
associated with high clustering and small characteristic path
length (diameter). Although in these networks most nodes
are not adjacent, they tend to share neighbors. In a similar
vein, most nodes in scale-free networks also follow this
characteristic. A small number of nodes, referred to as hubs,
show high connectivity. This property is consistent with the
power-law degree distribution.

Illicit supply-chain networks are not typically represented
by ER networks. Instead, Kinsella (2008) and Malm and
Bichler (2011) mentioned the scale-free structure of illicit
arms and narcotics markets. Tsirogiannis and Tsirogiannis
(2016), on the other hand, found conformity of both scale-
free and small-world characteristics in antiquity trafficking
networks. However, they refrained from confirming it as
scale-free, citing an insufficient number of nodes. Another
concept of criminal networks was provided by Borgatti and
Everett (2000), who split it into two node sets: core and per-
iphery. The former consists of densely interconnected nodes,
whereas the latter includes nodes that are sparsely con-
nected. In another sense, the core and periphery sets can be
considered as planners and executioners in the network,
respectively. Gimenez-Salinas Framis (2011) found evidence
of this structure in multiple Spanish cocaine traffick-
ing networks.

Recent advancements in network science have prompted
the adoption of complex networks in diverse disciplines.
One of its variants, multiplex or multi-relational networks,
has sparked interest among researchers working on illicit
trade. Bianconi (2013) defined multiplex networks as the
aggregation of networks with identical nodes but different
links. Each constituent network, i.e., layer or plex, specifies
the base of connectivity between actors (e.g., monetary, kin-
ship, resource). Incorporation of relational multiplexity
allows analysis from multiple perspectives, resulting in a bet-
ter understanding of illicit networks (Bright, 2015). Section
5.2 provides further discussion regarding this point.
Bahulkar et al. (2018a) and Baycik et al. (2018), on the other
hand, adopted interdependent network structures. Duijn
et al. (2014) offered a two-tiered drug trafficking network.
The first tier represented individual networks operating in a
particular market, whereas the second indicated a combin-
ation of all such networks. Bichler et al. (2017) termed them
as group and market structure, respectively. The representa-
tions mentioned above are quite similar in nature. In fact,
they all fall under the realm of multilayered networks. For
interested readers, we refer to the review of Kivel€a et al.
(2014), which provides a more precise definition of
these networks.

A variety of metrics are available in the literature for
explaining different aspects of a network, the simplest ones
being node and edge frequency. Perera et al. (2017) defined
their implications in supply chains. Bichler et al. (2017), on

the other hand, provided a list of metrics used in structural
analysis of drug supply networks. In the current article, we
will limit our discussion to the metrics separating illicit net-
works from others. Agreste et al. (2016) found the average
clustering coefficient in mafia networks to be higher than
that in social networks. Xu and Chen (2008) differentiated
drug trafficking networks from terrorist networks in terms
of higher path length, clustering coefficient, and efficiency
measures. In general, illicit networks are characterized by
low density and high centralization (Baker and Faulkner,
1993; Morselli, 2009a; Bright et al., 2012). Some, however,
disagree with this notion (Enders and Su, 2007). Bichler
et al. (2017) mentioned the centralization and density in
drug trafficking networks to be lower than legitimate and
co-offending networks, but greater than terrorist networks.

5.2. Ties in illicit network

Xu and Chen (2004) described organized criminal groups as
a set of offenders connected through various types of rela-
tionships. Factors facilitating these connections include trust,
triadic closure, roles in the supply chain, and network per-
formance (efficiency and security) (Morselli, 2009b; Bright
et al., 2019). They may also originate from already estab-
lished social networks of kinship, friendship, or ties
(McCarthy et al., 1998). Law enforcement agencies pursue
these links employing various techniques. However, it is
unlikely to discover all links within the network. Some of
the missing ties can be predicted using link analysis, which
will be discussed in Section 6.2.

While examining the associations among members in
illicit supply chains, one should acknowledge the diversity or
multiplexity present in them, which, until recently, was
ignored by traditional social network analysis (Everton,
2009; Papachristos and Smith, 2012). In the study of a drug
trafficking network by Bright et al. (2015), about 48% of the
nodes were involved in multiple types of relations.
Assuming the exchange of each particular resource as a dis-
tinct relationship category, they identified eight types of
links: drugs, money, precursors, premises, skills, informa-
tion, equipment, and labor. Krebs (2002) considered four
categories of criminal ties: trust, task, money and resource,
strategy and goal. Papachristos and Smith (2012) provided a
different classification scheme involving criminal, personal,
and legitimate contacts. None of these papers, however, con-
sidered adversarial relations representing rivalry or enmity.

Although incorporating multiplexity into networks does
provide for better analysis, considering too many categories
may also cause the constituent networks to be sparse and
incomplete. To avoid that, one might aggregate them into
groups or layers (Krebs, 2002). The question that arises then
is how many layers to use. Gera et al. (2017) related its
answer to the quality of community detection in the result-
ing networks. They also provided an index combining four
quality detection metrics referred to as: Normalized Mutual
Index, Purity, Rand Index, and Adjusted Rand Index.

10 R. ANZOOM ET AL.

5.3. Roles and positions in illicit network

Position or role in a network can be defined as the set of
nodes that are structurally substitutable (Xu and Chen,
2005). In the case of illicit networks, these are associated
with specific skills (i.e., human capital) and may vary
depending on activity or market niche (Bichler et al., 2017).
Reported indicators of complex roles include high degree
and betweenness centrality, as well as low clustering coeffi-
cient (Malm and Bichler, 2011; Calderoni, 2012). To identify
these roles, Xu and Chen (2005) suggested using positional
analysis, i.e., investigation of how similarly two nodes con-
nect to other network members.

Among the different types of roles and positions in net-
works, two are featured frequently in the literature: hubs and
brokers (Everett and Borgatti, 1999; Borgatti and Everett,
2006). Hubs maintain higher connectivity, whereas brokers
control the flow between different nodes. Kinsella (2008) pro-
vided a detailed classification of brokerage roles in the arms
trade. It identified a broker as a coordinator or liaison depend-
ing upon whether the connection is between members of the
same or different organizations, respectively. Based on the flow
direction, they again classified liaison into gatekeepers
(resource inflow) and representatives (resource outflow) Divi�ak
et al. (2019) provided another node classification scheme based
on position: visible, strategically positioned, marginal, and cen-
tral. We shall review it in Section 5.5.

5.4. Key nodes in illicit network

Gathering information on illicit networks is a difficult task.
With limited resources it becomes infeasible for law enforce-
ment agencies to surveil or disrupt the whole network.
Researchers, therefore, suggest focusing on key or central
nodes instead (Shaikh and Jiaxin, 2008). Several propositions
are available for their characterization in illicit networks.
Ballester et al. (2006) identified them as the ones with the
largest contribution to criminal activities. Baycik et al.
(2018) suggested targeting the highest-ranked criminals,
whereas Carley (2006) opted for removing emergent leaders.
Borgatti and Everett (2000) assumed key nodes to occupy
core positions in the network. However, this is not necessar-
ily the case since central actors might assume peripheral
positions to avoid detection (Baker and Faulkner, 1993;
Agreste et al., 2016). Schwartz and Rouselle (2009) suggested
two more perspectives for determining node importance:
human and social capital. Human capital denotes the posses-
sion of certain resources or skills, and associated nodes are
usually hard to replace (Robins, 2009). Based on this,
Hastings (2012) mentioned two types of actors in nuclear
trafficking networks: have and have-not. Social capital, on
the other hand, indicates the social connections with other
nodes. Nodes with high social capital (i.e., hubs and brokers)
are more capable of sharing information or resources with
others. For better identification, Bichler et al. (2017) sug-
gested using both human and social capital of nodes in the
investigation.

Centrality measures have been a popular choice among
researchers for identifying central nodes in illicit networks

(Bichler et al., 2017). Although several metrics exist (see Das
et al. (2018) for details), three have prevailed: degree, closeness,
and betweenness centrality (Shaikh and Jiaxin, 2008). The first
two are global network measures, whereas the last one is a
local measure. Patel et al. (2015) identified the key exporter
and importer in wildlife trafficking using out-degree and in-
degree centrality, while betweenness centrality designated top
intermediary countries. Kinsella (2008) used the same approach
to distinguish mediators in arms trades. Grassi et al. (2019)
used seven variants of betweenness centrality to identify crim-
inal leaders and found different results. They suggested choos-
ing a measure according to the nature of data, as since each
alternative captured discrete features of the nodes. Memon
(2012), on the other hand, recommended inclusion of tie
strength in centrality measurement.

Hussain and Ortiz-Arroyo (2008) applied Bayesian prob-
ability theory alongside social network analysis to detect key
nodes in criminal networks. Here, they ranked nodes based
on the change in entropy upon their removal from the net-
work. Farasat et al. (2016) provided two social network
approaches (hop count weighted and path salience) to iden-
tify high-valued terrorists from fused data. Taha and Yoo
(2016, 2017) provided two more approaches for deriving
prime nodes. The first one created a minimum spanning
tree of criminal networks from existing data. In that tree,
nodes received importance scores based on the number of
vertices that were dependent on them for existence. The
second method computed the relative individual influence of
nodes over the network. It aimed at capturing the immedi-
ate leaders of lower-level criminals.

While examining the positional importance of nodes
across different network layers, Bright (2015) found fluctua-
tions in the central position. No metric, however, was used
to resolve it. Toledo et al. (2020) applied node diversity ana-
lysis in a multiplex crime network to determine critical
nodes. They considered two types of diversity: connection of
node in various layers and heterogeneity of relationships in
the system.

Bichler et al. (2017) considered network topology and
detection algorithm to be determinants of quality of critical
node identification. Researchers usually test the effectiveness
of their approaches by comparing the result with a known
network. These measures are: sensitivity, specifity, accuracy,
recall, precision, F1 score, area under the ROC curve.

5.5. Network vulnerability, resilience and adaptability

A prerequisite for disrupting any network is to understand
its strengths and weaknesses. In the case of illicit networks,
these two are often intertwined. According to Bouchard
(2007), resilience has two features: resistance to disruption
and adaptation following the interruption. Cavallaro et al.
(2020) cited three factors governing these capabilities: struc-
ture, nodal position, and human capital. Scale-free and
multiplex networks contain structural redundancy, i.e.,
diversity of links, which grants them tolerance to random
disruption. However, this also makes critical nodes highly
susceptible to targeted disruption. Nodal positions can also

IISE TRANSACTIONS 11

inform about the vulnerability or resilience of a node.
Morselli (2010) used a scatter plot of betweenness against
degree centrality to describe this situation. Betweenness cen-
trality denoted the brokerage position of a node, whereas
degree centrality indicated its visibility or vulnerability. The
mean centrality scores divided the graph into four quad-
rants, which Divi�ak et al. (2019) mentioned as the location
for central, strategically positioned, marginal, and visible
nodes, respectively. Central nodes have high betweenness
and degree centrality. As a result, these nodes are valuable
and vulnerable to surveillance or interdiction by law
enforcement agencies. Strategically positioned nodes also
have high betweenness centrality, but lower visibility reduces
their detection probability. Bright (2015) assumed a node to
be strategically positioned if its betweenness and degree cen-
trality were at least one standard deviation above and below
the respective mean network score.

Besides resistance to disruption, illicit networks are also
capable of adapting and recovering from interruption,
increasing the number of steps for their destabilization. Basu
(2013) viewed their adaptation process as a cat and mouse
game. Law enforcement agencies try to put pressure on
illicit operations either by increasing surveillance or through
offensive actions. Traffickers respond by changing conceal-
ment techniques and transportation modes or shifting to
alternative routes. The latter phenomenon is known as the
balloon and cockroach effect, i.e., law enforcement actions
in one location results in the displacement of trafficking
activities to another location (Kleiman, 2011; Reuter, 2014).
Changes can occur in network size and reach as well. To
recognize these mutations, researchers use simulation
(Caulkins et al., 1993; Rydell et al., 1996; Dray et al., 2008;
Magliocca et al., 2019).

Networks often experience loss of active members due to
interdiction attempts by law enforcement agencies. In such
circumstances, one has to look for replacements. Trust plays
a crucial role in their selection process. Thus, substitutes
often come from short social distances from the cohesive
core (Duijn et al., 2014; Bright et al., 2017). However, for
replacing a specialist, one might have to go beyond the
existing network. Two significant risks arise here. First, it
may require cooperation with personnel whose reliability is
unknown. Second, the search increases network communica-
tion, raising network visibility as well as vulnerability
(Lindelauf et al., 2009). Duijn et al. (2014) provided a quan-
titative model for explaining the recovery mechanism in
illicit networks. According to it, when a particular node is
removed from the network, its adjacent (orphan) nodes look
to recover the link by connecting to replacement nodes.
These replacement nodes are selected by three methods: ran-
dom recovery, preference by distance recovery, and prefer-
ence by degree recovery. The first mechanism does not have
any preference and treats all nodes equally. The second pro-
cedure evaluates the distance of candidate nodes from the
orphan actor for selection, while the last one judges on their
degree centrality. Interestingly, the network came out to be
more efficient than before in all three mechanisms.
However, for that, they had to sacrifice their security,

leading to more scope for interdiction. In case of repetitive
disruption, the network eventually becomes completely vis-
ible, sinking its adaptation capability and causing failure.

5.6. Network dynamics

Carley et al. (2002) stated that failure to account for net-
work dynamics and adaptation can lead to erroneous policy-
making. However, not many studies are available concerning
the dynamics of illicit networks. One significant reason
behind this is the static nature of the acquired data, which
we discussed in Section 3.5. Nevertheless, this highly-antici-
pated field, often dubbed as the holy grail of network ana-
lysis, has been receiving increased attention over the last few
years. The prime objective of these studies is to comprehend
how different aspects of illicit networks change over time.
Bright and Delaney (2013), for example, discussed the struc-
tural changes that took place in a drug trafficking network.
In their study, centralization of an organization increased
when it became more profit-oriented, indicating a preference
for efficiency over security. However, this does not necessar-
ily imply the disregard of security. Bright et al. (2019) men-
tioned that actors optimized security through triadic closure,
trust, and communication through brokers. They based the
conclusion on a 2-year study of a drug trafficking organiza-
tion that evolved from a small social network into a large
profit-oriented corporation. A stochastic actor-oriented
model was applied to understand the formation of links,
whereas the Jaccard index was used to identify network sta-
bility over time. Recently, Meneghini et al. (2020) con-
structed a dynamic transnational cigarette trafficking
network that reflected its activity from 2008 to 2017.
Broccatelli et al. (2016), on the other hand, suggested bi-
dynamic line graph representation of illicit networks and
demonstrated its application on three case studies.

6. Uncovering illicit supply chains

Sections 4 and 5 provided a detailed review on the structure
and operations of the illicit supply chain and methods to char-
acterize them. However, different concealment and evasion
strategies by criminals create difficulty in the identification of
the supply chain. Researchers nevertheless endeavor to achieve
greater success in exposing them. Their efforts concentrate on
the detection of three principal entities: products, activities, and
networks. Section 4 has already addressed the second issue.
Therefore, this section will discuss the remaining two topics.

6.1. Detection of products

Land and sea ports of entry are regularly used by criminals
to smuggle illicit products across borders. Port authorities
try to identify these products using different inspection
measures. Although manual unpacking is the surest way to
identify illicit products, resource and budget restrictions
impede its complete implementation (Martonosi et al.,
2007). Therefore, the detection system typically follows a
risk-based approach instead. Generally, shipments are

12 R. ANZOOM ET AL.

assigned to different risk categories based on available infor-
mation and additional intelligence. The higher risk contain-
ers pass through subsequent detectors for further inspection.
If this increases the risk, the cargo is unpacked and physic-
ally checked. Otherwise, it is released. Decisions involved in
the design of an inspection system include the choice of
inspection strategies and tools, the number of detectors to
use, their arrangement, and the conditions for further
screening. And the system performance is assessed through
the detection rate and efficiency. Intuitively, one can
increase the detection rate by adopting a stringent inspec-
tion policy. However, this also increases lead time and cost,
reducing operational efficiency in turn. Thus, inspection
strategies have to be devised in a way that satisfies these
conflicting objectives.

The first issue to examine is the choice of inspection
methodology. Kantor and Boros (2010) described three
major inspection categories: document screening, scanning
test, and manual unpacking. Among these, document
screening is the least expensive. It attempts to detect docu-
ment fraud by analyzing patterns in trade (Hua et al., 2006;
Digiampietri et al., 2008; Yaqin and Yuming, 2010) or itin-
erary (Camossi et al., 2012; Dimitrova et al., 2014). Triepels
et al. (2018) included both approaches in their analysis,
which employed a Bayesian network model to predict the
presence of goods in the container. The second method,
scanning test, employs Non-Intrusive Inspection (NII)
equipment (e.g., Radiation Portal Monitor (RPM), gamma-
ray scanners) to examine the consignments. Since these
technologies need to be purchased and maintained, cost for
this method exceeds the previous one. Dimitrov et al. (2011)
mentioned several factors affecting the performance of the
detection equipment (e.g., material and container type, geo-
metric attenuation, shielding, background, sensing time).
Gaukler et al. (2012) evaluated three schemes for detecting
nuclear materials in the cargo: the existing Automated
Targeted System (ATS), radiography-based Hardness
Control System, and Hybrid Inspection System. In terms of
maximizing the minimum detection probability, the latter
two outperformed ATS under a wide range of conditions.
However, the decision was highly dependent on the reliabil-
ity of available data. Apart from this, operating costs and
the rate of false alarm are important as well. Especially, the
latter issue is quite significant since it can cause user dissat-
isfaction and port congestion, leading to lower competitive-
ness. Bak ir (2008) suggested comparing these issues against
the perceived security risk. He also advised against invest-
ment in new technologies unless the security risk becomes
high. Kantor and Boros (2010) developed an index to exam-
ine the cost-effectiveness of different inspection strategies. It
incorporated four factors: sensitivity, specificity, cost, and
detection rate.

Bakker et al. (2020) stressed on having more assets rather
than having better assets. This raises another salient ques-
tion: how many detectors to use in the system. Intuitively,
one would think of increasing the number as much as pos-
sible, as it would increase the detection rate. However, cost-
efficiency also needs to be considered in this decision.

Jacobson et al. (2006) compared the cost-effectiveness of
using single and two-device explosive detection architectures
at airports. Their analysis recognized the single device archi-
tecture as the better performer. Kretschmann and
M€unsterberg (2017), on the other hand, used a discrete-
event simulation framework to evaluate different combina-
tions of NII technologies in a border crossing. Here, they
identified the existence of a tradeoff between Type I (false-
clear) and Type II (false-alarm) errors in increasing identifi-
cation devices. Besides, adding more detectors did not
necessarily increase lead time and utilization rate. Instead, it
depended on the logical structure of the detection architec-
ture. Wein et al. (2007) optimized the number of RPMs in a
port considering budget capacity as well as queue length
and time.

Researchers have also investigated the location strategies
for detectors, often using stochastic network interdiction
models (Pan et al., 2003; Nehme, 2009; Dimitrov et al.,
2011). Here, the smuggler selects the path maximizing the
probability of undetected transportation, whereas the
authority (interdictor) installs detectors to minimize this
probability. Stochasticity arises in smuggler’s origin and des-
tination, type of product smuggled, detector probability, and
manner of shielding. Nehme (2009) provided three versions
of this problem: Stackelberg game, Cournot game, and a
hybrid game. In the first game, the smuggler is aware of the
detector locations. The second game allows both the inter-
dictor and the smuggler to act simultaneously. In the third
game, the smuggler knows some of the detector locations.
All three problems were formulated as mixed-integer pro-
gramming models and solved using branch-and-bound algo-
rithms. Wein et al. (2007) also optimized the spatial
positioning of RPMs in three designs of border inspection
systems. For that, they used a mathematical model consist-
ing of a queuing model, a detection model, and a cost
model. The objective was to minimize the mean detection
threshold for 95% detection probability under budget con-
straints. Under the operating assumption that the position-
ing of weapons within containers affects the detection rate,
both best-case and worst-case placement scenarios were con-
sidered. The optimal design reduced the existing detection
threshold to one-third, which was effective in detecting plu-
tonium, but not uranium.

As previously mentioned, inspection processes may
involve multiple steps, where the outcome of a particular
stage determines whether to elevate a consignment to its
subsequent stage or not. Since each stage might consist of
multiple detectors, one has to decide on the number of
device alarms that will lead to further screening. McLay and
Dreiding (2012) formulated this as a knapsack problem
model (Multilevel Knapsack Screening Problem), aiming to
maximize the detection rate within a fixed budget. For com-
parison, they introduced another threshold-based model
(Multilevel Threshold Knapsack Screening Problem). The
analysis showed that threshold-based policy was not optimal
under all conditions, but its detection probability was near-
optimal. Boros et al. (2009), on the other hand, developed a
large-scale linear model to determine the optimal inspection

IISE TRANSACTIONS 13

sequence. Here, they considered multiple thresholds for each
sensor. However, some have cautioned against using a pure
strategy since adversaries can learn of this and take adaptive
measures. An alternative can be using mixed strategies, i.e.,
random assignment of containers to different paths through
multiple detectors, which maintains constant risk for smug-
glers (Boros et al., 2009; Kantor and Boros, 2010). Apart
from these, Sherman et al. (2012) used a set of simulation
tools (Scenario Analysis, Decision Trees, Monte Carlo
Simulation) to identify the most cost-efficient inspection
rates. Extensive research also exists in design and analysis of
aviation security systems, which is summarized in the survey
by Lee et al. (2008) and Albert et al. (2021). Although the
papers focused mostly on the passenger transportation sys-
tem, they could also be applicable to freight transportation
and other transportation modes.

6.2. Identification of illicit networks

Section 5 presented different properties of known illicit net-
works and their analyses. In this section, our focus shifts
toward the discovery of these networks. Existing literature
offer multiple techniques to facilitate this endeavor. For
example, Diesner and Carley (2004) combined network text
analysis (Popping, 2000) with meta-matrix modeling to
obtain the representation of a Middle-Eastern illicit network.
Anwar and Abulaish (2014), on the other hand, used a
social graph-based text mining framework (n-gram tech-
nique and hyperlink-induced topic search) on criminal chat
logs to derive their network. Ozgul et al. (2012) developed
four network detection models that looked for node similar-
ity in different sets of features (e.g., crime location, date,
modus operandi, surname, hometown, co-offending data).
Testing the models on several databases, they observed that
narcotics trafficking network was the hardest to detect.

One might also be interested in the discovery of a par-
ticular element of the network rather than the whole. In that
sense, the detection techniques fall under three categories:
nodes, edges, and communities. The following subsections
provide a brief discussion on these issues.

Node detection: Members of illicit networks tend to exhibit
interactions that are suspicious and different from their
peers (Chandola et al., 2009). In other words, they are con-
sidered anomalies in a social network. Bindu et al. (2017)
applied an unsupervised anomaly detection method in a
multilayered social network to identify such anomalies. In
that paper, they defined nodes with near-star or clique
neighborhood topology as anomalous. Suehr and Vogiatzis
(2018), on the other hand, used integer programming and a
k-club path-like formulation to identify nodes that might
misrepresent themselves in social media. In this case, the
network was single-layered.

Link analysis: Link analysis/prediction involves estimating
the likelihood of a link existing between two nodes based on
observed ties and attributes of the nodes. Researchers have
proposed different algorithms over time for this purpose

(see Al Hasan and Zaki (2011) and L€u and Zhou (2011) for
details). However, our focus is limited to the studies in illicit
(and criminal) networks. Schroeder et al. (2007) discussed
four principal methods for criminal link analysis: heuristics-
based, template-based, similarity-based, and statistical. Link
detection in these methods depend on decision rules, pre-
defined template, similarity between entities, and lexical sta-
tistics, respectively.

Xu and Chen (2004) proposed two shortest path algo-
rithms (priority-First-Search (PFS) and two-tree PFS) to
identify associations between criminal entities. Logarithmic
transformation converted the link weights into distances.
When applied to a drug network, PFS fared better in terms
of execution time. Schroeder et al. (2007) combined co-
occurrence analysis, shortest path algorithm, and a heuristic
approach for automatic link analysis. Isah et al. (2015) used
a bipartite network model to infer hidden ties between
actors in the illicit medicine supply chain. Initially, they con-
structed a network with two node sets: actors and resources.
Later, it got converted into an actor–actor network, where
an edge connected two nodes if they had a common neigh-
bor in the previous network. The model was validated using
a standard network algorithm for structural and community
analysis. Lim et al. (in press) compared a link prediction
model based on reinforcement learning against machine
learning methods. The former model provided better accur-
acy with a smaller dataset. Marciani et al. (2017) investigated
different social network metrics for detecting and predicting
links in an evolving criminal network. Moreover, the appli-
cation of data stream processing approach allowed the
extraction of valuable information in real-time. For that,
they also introduced three similarity social network metrics.
Calderoni et al. (2020) applied several link prediction algo-
rithms on a mafia network and observed that algorithms
using the full graph topology had better accuracy and con-
sistency. They also investigated the impact of different data
attributes on the prediction accuracy. Results from simula-
tion showed that the link predictions remain robust as long
as information on the network was fairly complete, and the
unobserved edges followed a generative law.

Community detection methods: Communities, also known
as clusters or modules, are independent compartments in a
network that exhibit a higher concentration of edges within
themselves than with members of other sections (Fortunato,
2010). Calderoni and Piccardi (2014) mention communities
as a natural phenomenon in illicit networks. Thus, their dis-
covery can provide useful insights regarding the network
structure. Xu and Chen (2004) used a hierarchical clustering
approach for automatic detection of subgroups in criminal
networks. For this purpose, they used a Reciprocal Nearest
Neighbor-based complete-link algorithm. Isah et al. (2015)
applied four community detection algorithms (Girvan
Newmann, Clauset Newmann Moore, Wakita Tsurumi,
Walktrap) to a network of medicine counterfeiters. Neither
of the aforementioned approaches allowed the presence of a
particular node in multiple groups. Robinson and Scogings
(2018) and Zhao et al. (2020) included this provision in

14 R. ANZOOM ET AL.

their works. The former developed a novel graph mining
method (GraphExtract algorithm) that detected the sub-
graphs of entities involved in atomic criminal events. Zhao
et al. (2020), on the other hand, provided a matrix factoriza-
tion method for detecting drug vendor communities across
three social platforms. Besides these, Calderoni and Piccardi
(2014) applied local and global (max modularity) commu-
nity detection approaches to investigate clustering in mafia
networks. For better performance, Bahulkar et al. (2018b)
suggested predicting hidden links first and then performing
community detection on the augmented network. Recently,
Sangkaran et al. (2020) surveyed graph-analytics-based com-
munity detection methods used in criminal networks. Apart
from that, one may also look into algorithms used in other
domains (see works by Parthasarathy et al. (2011) and
Malliaros and Vazirgiannis (2013)).

7. Disruption of illicit trade

The key to subdue illicit trades lies in the disruption of their
associated networks. According to Carley et al. (2003), net-
work disruption is the act of reducing, or possibly eliminat-
ing a network’s capability to disseminate resources
efficiently. Usual targets for such interruption are the peo-
ple, activity, or assets in the network (Bright, 2015; Bright
et al., 2017). Rydell et al. (1996) discussed four approaches
to intervene in the narcotics trade: source-country control,
interdiction, domestic enforcement, and treatment. The first
three methods focused on curbing the supply, whereas the
last one concentrated on repressing demand. Clifton and
Rastogi (2016), on the other hand, discussed four social net-
work intervention strategies: individual, segmentation,
induction, and alteration. The majority of these strategies
were designed from the perspective of law enforcement
agencies; however, some are available for business organiza-
tions as well. Capitalizing on these techniques, different
institutions try to confront illicit trade.

However, disrupting illicit trades is not easy. The first
difficulty arises in their detection due to various conceal-
ment measures. Second, even after disruption, networks can
spring back through adaptive measures or spread to new
regions. Third, actions to disrupt illicit trades can hurt per-
formances of licit trades as well (Cedillo-Campos et al.,
2014). These have led some researchers to question the
rationale for disruptive measures, especially in drugs (Duijn
et al., 2014; Magliocca et al., 2019). The proponents of inter-
diction, however, stress upon its symbolic and moral value.
In their view, the current interdiction approach is inad-
equate, not ineffective. And this inefficiency stems from
resource limitations, inadequate coordination, and corrup-
tion (Basu, 2014a). Considering both perspectives, we have
divided the disruption approaches of illicit trades into two
broad categories: invasive and noninvasive. Invasive
approaches involve a direct engagement with the illicit net-
work stakeholders. Noninvasive approaches look for alter-
nate efforts to influence the trade indirectly. The following
sections discuss both approaches in detail, along with the
methodologies involved.

7.1. Measurement of disruption

Researchers have suggested different measures to evaluate
disruption in illicit trade and its associated networks.
Traditional scales for illicit trade disruption include the
change in number of retailers, sales volume, and retail price
(Caulkins and Padman, 1993; Crane and Rivolo, 1997;
Caulkins and Hao, 2008). Schneider (2008) added several
indicators specific to fencing: change in the market, disposal
time, and burglary frequency. In the network interdiction
literature, two poplar measures for disruption are the reduc-
tion in flow or increase in cost. Apart from these, several
network-based metrics have been used by researchers,
including the average path length, degree centralization,
number of connected components, size of the largest con-
nected component (Tsvetovat and Carley, 2003; Agreste
et al., 2016). Patel et al. (2015) suggested two new measures
in wildlife trade disruption: fragmentation index and
weighted reach index. The former denoted the proportion of
nodes isolated after the removal of interdicted nodes,
whereas the latter represented the weighted distance of non-
key nodes to key nodes.

7.2. Invasive disruption techniques

An invasive approach to disrupt an illicit supply chain typic-
ally involves law enforcement agencies adopting an offensive
stance against the illicit supply chain members. Three major
disruption approaches are found in the literature: social net-
work analytic methods, network interdiction, and crack-
down. Each approach is discussed separately in the
following subsections.

Social network analytic methods: Section 5.4 discussed the
significance of key nodes in network disruption as well as
the techniques to identify them. In this segment, we intend
to address their removal strategies. Generally, there are two
ways of removing nodes in a network: random or targeted.
Random removal, in essence, is not a concerted strategy. It
resembles opportunistic law enforcement interventions
against criminal networks, e.g., stop and search, vehicle
stops, opportunistic seizures (Bright et al., 2017). Targeted
removal, as the name suggests, attacks nodes with specific
attributes and with a certain goal in mind. Bright et al.
(2017) discussed five targeting methods: three focused on
social capital (based on betweenness, degree, cut-set), and
two focused on human capital (possession of money and
precursor chemicals). Results from simulation identified
betweenness and money as the best attributes to use. Duijn
et al. (2014) presented two new disruption strategies for the
cannabis supply chain: value chain degree and specific value
chain role.

Agreste et al. (2016), on the other hand, classified disrup-
tion strategies as parallel or serial, depending on whether
they are removed simultaneously or sequentially. Although
serial interdiction performed better in their experiment, it
may allow the network to reorganize between successive
attacks. From this perspective, the authors found parallel
interdiction to be more realistic. Patel et al. (2015) noted

IISE TRANSACTIONS 15

that removing members based on traditional centrality
measures does not necessarily cause the highest disruption.
Instead they employed the key player problem approach
(Borgatti, 2006) to identify the set of optimal nodes whose
removal maximized the fragmentation in wildlife trafficking
network, increasing the number of connections required to
travel from one node to another.

Network interdiction: Section 6.1 introduced network inter-
diction for detecting illicit products. In this section, we dis-
cuss further its application in illicit network disruption.
Research on network interdiction perhaps started with
Wollmer (1964) who attempted to identify the best set of
arcs to remove. Later, McMasters and Mustin (1970) intro-
duced a model to interdict the opponent’s supply network
subject to budget constraints. Over time, the literature has
been enriched with multiple studies (see, e.g., the recent sur-
vey of Smith and Song (2020)). In a recent study on illicit
supply chains, Jabarzare et al. (2020) categorized network
interdiction methods into four major categories: shortest
path, facility assignment, minimum cost flow, and maximum
flow. The last category has been mostly applied to disrupt
illicit supply chains. Wood (1993) developed a deterministic
network interdiction model for analyzing the actions against
narcotics flow across South America and showed it to be
NP-complete. Washburn and Wood (1995) formulated the
problem as a two-person zero-sum game. Meng (2013) opti-
mized border patrol routes using a strictly mixed strategy
Nash equilibrium in a two-player game. Guo et al. (2016)
introduced a novel Stackelberg game model, and proposed a
column and constraint generation algorithm to solve it.
Jabarzare et al. (2020), on the other hand, studied the
dynamic maximum flow interdiction of an illicit supply net-
work with multiple commodities, sources, and sinks. Here,
they proposed two reformulations of the min–max bi-level
mathematical model into a mixed-integer model, and pro-
vided a solution method based on the Benders decompos-
ition and different accelerating strategies (e.g., Super Valid
Inequalities). Stochastic models have also been proposed by
researchers with frequent application in nuclear material
detection systems (Pan et al., 2003; Morton et al., 2007). In
addition, Zhang et al. (2018) applied a stochastic shortest
path network interdiction model to a network of illicit path-
ways along the Arizona–Mexico border. Sadeghi and Seifi
(2019) applied a two-stage maximum flow network interdic-
tion problem with endogenous uncertainty. Tezcan and
Maass (2020) used single and multi-stage stochastic network
interdiction models to tackle human trafficking. Recently,
researchers have also attempted to represent and interdict
illicit networks as complex networks. Baycik et al. (2018)
applied network interdiction in a system comprising two
interdependent networks: physical and information. The tar-
get was to minimize the maximum flow of physical prod-
ucts, and novel multi-step dual-based reformulation
technique was developed to solve it. Inspired by this,
Bahulkar et al. (2018a) proposed a framework for interdict-
ing three interdependent networks: smuggling, money, and
money laundering.

Crackdown: Crackdown denotes the sudden intensification
in law enforcement activities in a particular area to increase
the perceived or actual threat of apprehension among crimi-
nals (Davis and Lurigio, 1996). In illicit supply chains, its
application is mostly found in the retail stage. Caulkins
(1993) introduced a model to describe the impact of crack-
down on the narcotics market. That paper also addressed
questions regarding the features of an optimal crackdown
strategy (e.g., target selection, enforcement level). Based on
his analysis, Caulkins suggested targeting one market at a
time and identified the threshold enforcement level to col-
lapse the market. However, some may intend to minimize
the market size rather than causing complete collapse. In
such cases, the market may attempt to recuperate. To avoid
this, Caulkins prescribed maintenance measures as a follow-
up to the disruption. The resource required in this phase is
much less than the disruption phase and depends on the
number of dealers at the end of the disruption (Baveja et al.,
2000). Baveja et al. (1993) introduced two dynamic enforce-
ment policies: the first policy decreased the enforcement
level gradually over time, whereas the second strategy mod-
eled the enforcement level as a function of the number of
dealers in the market. Market size and characteristics also
play a significant role in determining the crackdown policy
(Baveja et al., 1993; Kort et al., 1998). Kort et al. (1998)
developed an optimal control model to identify the crack-
down rate in three market settings: buyers, sellers, and
mixed market. Using the maximum principle, they showed
that it was easier to disrupt a sellers’ market than a buyers’
market. For the mixed scenario, the fate of the market
depended on its initial size. If sufficiently small, the market
collapses; otherwise, the number of dealers converges to a
stable saddle point equilibrium. Baveja et al. (2000) devel-
oped another optimal control model consisting of two states
(number of users and budget) and one control variable
(enforcement). The authors recommended tailoring enforce-
ment policy according to the risk nature of dealers. They
also assumed that the outcome of the crackdown should be
apparent within a short time, such as a week.

7.3. Noninvasive disruption approach

In addition to invasive strategies, governments are also look-
ing into alternate policies that can hurt illicit trades.
Moreover, licit organizations affected by illicit trades can
also play a part in disruption, especially in the case of coun-
terfeiting. These have given rise to different noninvasive
approaches over the last few years. Among these, marketing
strategy is frequently used by government authorities and
licit organizations to create awareness among consumers
against illicit trades (Schneider, 2008; Chaloupka et al.,
2015). Konrad et al. (2017) suggested optimizing media
investment to maximize awareness about human trafficking.
Such marketing approaches are also applicable to illicit trad-
ers. Patel et al. (2015), for instance, suggested disseminating
educational messages to encourage refraining from wildlife
trafficking. Another interesting idea is the optimization of

16 R. ANZOOM ET AL.

sentencing policy to deter people from participating in illicit
trades (Caulkins and Padman, 1993).

To tackle counterfeiting, a brand company can also adopt
strategies regarding pricing, quality, and technology: e.g.,
price rebates, quality enhancement, and R&D investments.
Cho et al. (2015) evaluated the effectiveness of these
approaches from three perspectives: profit of brand compa-
nies, profit of counterfeiters, and consumer welfare. They
also showed that the same strategy is not applicable for
countering both deceptive and non-deceptive counterfeiting.
Using an innovative idea, Kumar and Tripathi (2019) pro-
posed a block chain model for securing medicine supply
chain from counterfeits.

Often, the targeted network might be out of a country’s
jurisdiction. Shan and Zhuang (2015) presented a similar
scenario regarding the WMD supply chain where terrorists
plan to attack with weapons sourced from another criminal
organization. To repeal this threat, the country at risk can
reach out to the supplier’s native country with an offer to
subsidize the market disruption. The collaboration within
the governments and criminal groups was determined by
the amount of subsidy and payoff, along with additional var-
iables (e.g., attacking cost, preparation cost, proliferation
cost). Using a game-theory approach, the authors modeled
their interactions as two individual sub-games (proliferation
and subsidization) and integrated them into a four-player
game. Under different combinations of the variable states,
the model provided the optimal policies to follow.

7.4. Scheduling of disruption

Naik et al. (1996) introduced the Crackdown Scheduling
Problem (CSP), an analytical framework to identify the opti-
mal sequence of crackdowns on drug markets. Two phases
were considered: crackdown and maintenance. An exponen-
tial-time algorithm was employed to solve the general case
of the problem. They also provided an approximation algo-
rithm to solve a specific scenario. Cai et al. (1998) developed
a quasi-polynomial time approximation algorithm to solve
the CSP with a monomial cost function. Baveja et al. (2004)
developed a sequential crackdown model where enforcement
decisions were made daily. The model used a probabilistic
framework, where the probabilities of drug dealing and
dealer incarceration depended on several factors. The opti-
mal strategy was a cyclic one, which resembled a crack-
down-backoff strategy.

Malaviya et al. (2012) developed a multi-period network
interdiction problem that focused on scheduling the activ-
ities of law enforcement for successfully interdicting crimi-
nals in a narcotics supply chain. Enayaty-Ahangar et al.
(2019) solved this problem with a logic-based decomposition
approach along with constraint programming.

7.5. Disruption policies: Evaluation and selection

With the increase in the number of intervention strategies,
it is reasonable to examine their efficacy. Rydell et al. (1996)
assessed the cost-efficiency of four narcotics control

strategies and found addiction treatment to be the best
choice. Mazerolle et al. (2007) conducted a systematic review
of five drug enforcement policies. Their results found pro-
active interventions involving partnerships between the
police and third parties rendering the highest disruption.
Kovari and Pruyt (2012) used a system dynamics simulation
model to provide insights regarding the effects of proposed
policies to combat human trafficking. Data envelopment
analysis can be another possible mechanism.

It is also possible to adopt multiple strategies simultan-
eously. For example, in narcotics control, efforts are
required in both interdicting the supply as well as the treat-
ment of the users or victims. Questions may arise regarding
how to allocate budget to these policies. Tragler et al. (2001)
modeled this dilemma as an optimal control problem with
the objective to minimize social costs due to drug use and
control. Their solution implied that if the problem was in its
early stage, then it would be optimal to invest highly in both
enforcement and treatment to possibly eradicate it.
Otherwise, they suggested moderating it initially with
enforcement and later with a gradual increase in treatment
control. Recently, Baycik et al. (2020) proposed a Markov
Decision Process framework to analyze a resource allocation
problem for law enforcement that aimed to balance intelli-
gence and interdiction decisions to combat narcotics traf-
ficking. They applied a column generation technique and a
heuristic to solve this problem.

7.6. Coordination in disruption

Section 3.5 discussed the importance of coordination among
different organizations in combating illicit trade. However,
research in this domain is still relatively unexplored. Using
data envelopment analysis and cooperative game theory,
Lozano (2012) showed that data sharing among different
organizations improves cost-efficiency for everyone,
although not uniformly. Sharkey et al. (2015) considered
three coordination settings: centralized, decentralized, and
information-sharing. The first setting involves comprehen-
sive planning by a single central authority. In the second
environment, each agency decides on its own without con-
sideration of others. The third scenario also provides inde-
pendence in decision making but requires disclosure of the
decisions to others. Sharkey noted centralized environment
as the ideal state, yet backed the information-sharing envir-
onment under realistic conditions.

Wilt and Sharkey (2019) developed a maximum flow net-
work interdiction model to quantitatively assess the impact
of coordination in illicit supply chain interdiction. Their
study identified that the importance of coordination
increases when the network gets sparse. Moreover, return
(disruption inflicted) on investment was greater for a coor-
dinated than for an uncoordinated environment. This result,
however, was observed for a smaller budget level. The model
was applied to a five-tiered drug supply chain for identifying
the optimal configuration of coordination. Out of six combi-
nations between five agencies, the one between federal, state,
and municipal law provided the best result.

IISE TRANSACTIONS 17

8. Research gap and opportunities

The previous seven sections presented and discussed differ-
ent aspects of illicit trades and relevant literature. While
reviewing them, we identified several research gaps. These
are not intended to be comprehensive; instead, they will
serve to thwart illicit trade through IE/OR methods and
hopefully advance the state-of-knowledge of the latter. The
following paragraphs expand on and summarize them with
a hope to offer potential directions.

8.1. Illicit trade activity index

Although all sorts of illicit trade count toward criminal
offense, it is not practical for law enforcement agencies to
treat each of them with the same urgency or intensity.
Therefore, it becomes essential to critically examine the
trades from multiple perspectives before committing resour-
ces to their disruption. One possible scheme involves com-
paring the impacts of illicit businesses on a specific sector
(e.g., society, economy) or across all streams. Once the
desired number of significant categories are selected, focus
should shift toward identifying the factors that most affect
their performance. Quantifying their influence could lead to
the development of a vulnerability index, which implies the
proliferation risk of illicit trades in a particular region.
Although there exists one such index in the literature, it
does not differentiate between different trade categories. In
addition, its calculation involves a simple weighted average
of factor scores based on subjective opinions. In reality,
illicit trades are opportunistic crimes, and their growth,
extent, and impact are dynamic with time and dependent
upon the prevailing environmental conditions. There may
also exist connections between prevalence of different trade
categories (e.g., human traffickers may use their victims for
smuggling drugs). Efforts are needed to characterize the
influence of these conditioning factors on each trade, prefer-
ably through complex, probabilistic, and data-driven model-
ing. The resultant models may be of great value to policy-
makers, assessing the efficacy of different strategies. They
might also assist in quantitative assessment of illicit supply
chain performance, which has not received much attention
in the past. Development of performance metrics for the
whole supply chain and its drivers is a potential task for the
future. It is also worth investigating whether indices in licit
supply chain literature apply to the illicit one. Research
advancement in this arena is expected to improve the design
of supply chain disruption strategies. Finally, it is worth
mentioning that research on illegal supply chains is biased
toward a select few trade categories (e.g., contraband and
counterfeit by trade characteristics, narcotics by product).
Initiatives are required in the future to mitigate this gap and
ensure uniform progress in research across all categories.

8.2. Data source management and quality assessment

Since data on illicit trades are subject to a variety of short-
comings, it might be a good idea to assess their quality

before employing them for decision-making. The evaluation
can encompass multiple criteria (e.g., provenance, complete-
ness, dynamics) depending on the research objective. This
should also lead to the discovery of relationships between
data quality and performance of different analytic methods.
Implementation of data imputation techniques might be one
way to alleviate some deficiencies, i.e., incompleteness.
Another approach involves the generation of synthetic net-
works that resemble the characteristics of real-life illicit net-
works (Elsisy et al., 2020). One can also make the data
accumulation process more efficient by developing a shared
database. However, its design framework must address secur-
ity concerns, i.e., protection from adversarial interception
(corrupt agents and illicit network members). Fusion of het-
erogeneous data sources is another possible research direction
for the future. The classic tradeoff between investing resour-
ces to acquire new data for improved decision-making and
the interdiction process itself remains relevant, and modeling
and optimization methods can be advanced further.

8.3. Complex network analysis with rivalry

Analysis of illicit supply networks hitherto has been limited
to a few familiar models. Recent developments in network
science have brought forward further novel and intricate
network structures (Kivel€a et al., 2014). Applicability of
these models is a possible avenue to explore. Moreover,
most of the analyses have focused on a single organizational
network, whereas in reality, there might be multiple.
Excluding occasional truces, these organizations naturally
compete with each other, denoting adversarial relations
between their members. It is also possible to have rivalry
within the same organization. Consideration of these ties
can have paradoxical implications for law enforcement agen-
cies. On one end, rivalry reduces the resources of the net-
work members, making them easier to interdict. However,
disrupting one party would not necessarily eliminate/weaken
the trade since it augments the power (trade volume) of its
rivals. In such contexts, one would need to rethink network
analysis and disruption strategies (critical node identifica-
tion, whether to attack one network or both, sequentially or
simultaneously). Multi-player game-theoretic models may
have potential application here. Implementation of social
network analytic techniques should also be worthwhile, as
long as it incorporates the additional network features in
its metrics.

It might also be beneficial for law enforcement agencies
to instill or intensify rivalry between a cluster of nodes in
illicit networks. The decision depends on the expected util-
ity, i.e., impact on network resilience and the cost involved.
If approved, the task would then be to identify the optimal
node-set for instilling this rivalry. One possible approach is
to target nodes performing similar roles. More advanced
community detection methods can be developed to assist
the search procedure, especially those that consider side con-
straints that reflect specific patterns based on information
from previously interdicted illicit supply chains. In the case
of rivalry, it is important that side constraints effectuate in-

18 R. ANZOOM ET AL.

fighting in the cluster and can cause a significant part of the
network to fail alongside. Moreover, dynamic network ana-
lysis could be employed for measuring network resilience.

8.4. Further concerns in detection and disruption

The last few decades have seen significant research advance-
ments in the detection and disruption of illicit supply
chains. However, there remain avenues to explore. One is to
incorporate both positive and negative consequences of dis-
ruption strategies in their design. Researchers also need to
consider the displacement of trades resulting from disrup-
tion. Pattern analysis and facility location models could be
useful for predicting these future destinations. Furthermore,
resources need to be deployed to these areas for arresting
the spread, adding complexity to the resource allocation
decision. We can also represent the problem through
dynamic network interdiction models, where the cost of
interdiction or movement across edges is variable over time.
It is also of interest to predict the life expectancy of illicit
markets through survival analysis.

Coordination and corruption play critical roles in illicit
supply chain disruptions. However, only a handful of stud-
ies have addressed them through modeling (Geller et al.,
2011; Wilt and Sharkey, 2019). Researchers should also put
more focus on noninvasive disruption strategies. For
instance, optimization of regional tax policy can contribute
to a decline in parallel imports. Redesign of incentive poli-
cies can also deter distributors and retailers from engage-
ment in illicit trades. As for detection of supply chain
entities, anomaly detection methods can be useful, especially
for exposing corrupt government agents as well as deceitful
suppliers and consumers. Furthermore, incorporation of
Natural Language Processing and image recognition meth-
ods has the potential to increase the detection rate of sales,
advertisement, or discussion regarding illicit products. A
particular challenge here is the accurate differentiation
between legal and illegal products, where AI/ML methods
are most befitting.

9. Conclusion

There is no question about the threat illicit trade poses to
our economy, society, and environment. Its defiant presence
despite disruptive measures is a matter of concern across the
whole world. The complex and diverse nature of illicit trade
occludes its detection and interpretation, resulting in limited
success from interruption. Overcoming this challenge
requires further participation from researchers across differ-
ent disciplines, especially from the field of operations
research and data analytics. Inspired by this, this article
attempted to provide a comrehensive review of illicit supply
chains. Here, we outlined the general concepts in illegal
trade and its supply chain/network, summarized the
advancements in research so far, and pointed out future
directions for research. However, by no means should this
review be considered exhaustive. The first reason is the
exclusion of virtual products and services from our

discussion. Second, due to space limitations, we were not
able to include all the works that are ongoing in this field.
Nevertheless, we are hopeful that our work will assist
researchers engaged in this topic as well as encourage others
to join the fight against these reprehensible activities.

Acknowledgments

The authors would like to thank two anonymous reviewers and the
Associate Editor for their helpful comments that have led to an
improved paper.

Notes on contributors

Rashid Anzoom is a PhD student in the Department of Industrial and
Enterprise Systems Engineering at University of Illinois, Urbana-
Champaign. He received his MSc (2019) and BSc (2017) degrees in
industrial and production engineering from Bangladesh University of
Engineering and Technology. Rashid’s research interest includes opera-
tions research, data analytics, and supply chain.

Rakesh Nagi is Donald Biggar Willett Professor of Engineering at the
University of Illinois, Urbana-Champaign. He served as the
Department Head of Industrial and Enterprise Systems Engineering
(2013-2019). He is an affiliate faculty in CS, ECE, CSL, and CSE.
Previously he served as the Chair (2006-2012) and Professor of
Industrial and Systems Engineering at the University at Buffalo
(SUNY) (1993-2013). He has more than 200 journal and conference
publications. Dr. Nagi’s academic interests are in big graphs/data, social
networks, GPU-accelerated computing, graph algorithms, production
systems, applied/military operations research and data fusion using
graph theoretic models.

Chrysafis Vogiatzis is a Teaching Assistant Professor in the
Department of Industrial and Enterprise Systems Engineering at the
University of Illinois, Urbana-Champaign. Previously he was an assist-
ant professor of Industrial and Systems Engineering at North Carolina
A&T State University. He received his PhD (2014) and MS (2012)
degrees in industrial and systems engineering at the University of
Florida, and his Dipl. Eng. (2009) degree in Electrical and Computer
Engineering at the Aristotle University of Thessaloniki in Greece. His
academic interests include network optimization and analysis, decom-
position techniques for combinatorial optimization, and applied opera-
tions research.

ORCID

Rashid Anzoom http://orcid.org/0000-0003-4699-6379
Rakesh Nagi http://orcid.org/0000-0003-4022-6277
Chrysafis Vogiatzis http://orcid.org/0000-0003-0787-9380

References

Agreste, S., Catanese, S., de Meo, P., Ferrara, E. and Fiumara, G.
(2016) Network structure and resilience of mafia syndicates.
Information Sciences, 351, 30–47.

Al Hasan, M. and Zaki, M.J. (2011) A survey of link prediction in
social networks, in Social Network Data Analytics, Springer, Boston,
MA, pp. 243–275.

Albert, L.A., Nikolaev, A., Lee, A.J., Fletcher, K. and Jacobson, S.H.
(2021) A review of risk-based security and its impact on TSA
PreCheck. IISE Transactions, 53, 657–670.

Albright, D., Brannan, P. and Stricker, A.S. (2010) Detecting and dis-
rupting illicit nuclear trade after A.Q. Khan. The Washington
Quarterly, 33, 85–106.

IISE TRANSACTIONS 19

Anwar, T. and Abulaish, M. (2014) A social graph based text mining
framework for chat log investigation. Digital Investigation, 11,
349–362.

Arroyave, F.J., Petersen, A.M., Jenkins, J. and Hurtado, R. (2020)
Multiplex networks reveal geographic constraints on illicit wildlife
trafficking. Applied Network Science, 5, 20.

Bahulkar, A., Baycik, N.O., Sharkey, T., Shen, Y., Szymanski, B. and
Wallace, W. (2018a) Integrative analytics for detecting and disrupting
transnational interdependent criminal smuggling, money, and money-
laundering networks, in 2018 IEEE International Symposium on
Technologies for Homeland Security., IEEE Press, Piscataway, NJ, pp. 1–6.

Bahulkar, A., Szymanski, B.K., Baycik, N.O. and Sharkey, T.C. (2018b)
Community detection with edge augmentation in criminal networks,
in 2018 IEEE/ACM International Conference on Advances in Social
Networks Analysis and Mining., IEEE Press, Piscataway, NJ, pp.
1168–1175.

Baker, W.E. and Faulkner, R.R. (1993) The social organization of con-
spiracy: Illegal networks in the heavy electrical equipment industry.
American Sociological Review, 58, 837–860.

Bak Ir, N.O. (2008) A decision tree model for evaluating countermeas-
ures to secure cargo at United States southwestern ports of entry.
Decision Analysis, 5, 230–248.

Bakker, C., Webster, J., Nowak, K.E., Chatterjee, S., Perkins, C.J. and
Brigantic, R. (2020) Multi-game modeling for counter-smuggling.
Reliability Engineering & System Safety, 200, 106958.

Ballester, C., Calv�o-Armengol, A. and Zenou, Y. (2006) Who’s who in
networks. Wanted: The key player. Econometrica, 74, 1403–1417.

Barab�asi, A.-L. and Albert, R. (1999) Emergence of scaling in random
networks. Science, 286, 509–512.

Basu, G. (2013) The role of transnational smuggling operations in illicit
supply chains. Journal of Transportation Security, 6, 315–328.

Basu, G. (2014a) Combating illicit trade and transnational smuggling:
Key challenges for customs and border control agencies. World
Customs Journal, 8, 15–26.

Basu, G. (2014b) Concealment, corruption, and evasion: A transaction
cost and case analysis of illicit supply chain activity. Journal of
Transportation Security, 7, 209–226.

Baveja, A., Batta, R., Caulkins, J.P. and Karwan, M.H. (1993) Modeling
the response of illicit drug markets to local enforcement. Socio-
Economic Planning Sciences, 27, 73–89.

Baveja, A., Feichtinger, G., Hartl, R., Haunschmied, J. and Kort, P.
(2000) A resource-constrained optimal control model for crackdown
on illicit drug markets. Journal of Mathematical Analysis and
Applications, 249, 53–79.

Baveja, A., Jamil, M. and Kushary, D. (2004) A sequential model for
cracking down on street markets for illicit drugs. Socio-Economic
Planning Sciences, 38, 7–41.

Baycik, N.O., Sharkey, T.C. and Rainwater, C.E. (2018) Interdicting lay-
ered physical and information flow networks. IISE Transactions, 50,
316–331.

Baycik, N.O., Sharkey, T.C. and Rainwater, C.E. (2020) A Markov deci-
sion process approach for balancing intelligence and interdiction
operations in city-level drug trafficking enforcement. Socio-Economic
Planning Sciences, 69, 100700.

Beckert, J. and Wehinger, F. (2013) In the shadow: Illegal markets and
economic sociology. Socio-Economic Review, 11, 5–30.

Becucci, S. (2004) Old and new actors in the Italian drug trade: Ethnic
succession or functional specialization? European Journal on
Criminal Policy and Research, 10, 257–283.

Bianconi, G. (2013) Statistical mechanics of multiplex networks:
Entropy and overlap. Physical Review. E, Statistical, Nonlinear, and
Soft Matter Physics, 87, 062806.

Bichler, G., Malm, A. and Cooper, T. (2017) Drug supply networks: A
systematic review of the organizational structure of illicit drug trade.
Crime Science, 6, 2.

Bindu, P., Thilagam, P.S. and Ahuja, D. (2017) Discovering suspicious
behavior in multilayer social networks. Computers in Human
Behavior, 73, 568–582.

Boivin, R. (2013) Drug trafficking networks in the world economy, in
Crime and Networks, Routledge New York, NY, pp. 12–194.

Borgatti, S.P. (2006) Identifying sets of key players in a social network.
Computational and Mathematical Organization Theory, 12, 21–34.

Borgatti, S.P. and Everett, M.G. (2000) Models of core/periphery struc-
tures. Social Networks, 21, 375–395.

Borgatti, S.P. and Everett, M.G. (2006) A graph-theoretic perspective
on centrality. Social Networks, 28, 466–484.

Boros, E., Fedzhora, L., Kantor, P., Saeger, K. and Stroud, P. (2009) A
large-scale linear programming model for finding optimal container
inspection strategies. Naval Research Logistics (NRL), 56, 404–420.

Bouchard, M. (2007) On the resilience of illegal drug markets. Global
Crime, 8, 325–344.

Bradshaw, H. (2016) Glowing pockets: Modeling illicit nuclear and
radiological trafficking networks in the former Soviet Union, in
2016 National Conference on Undergraduate Research.

Bright, D., Greenhill, C., Britz, T., Ritter, A. and Morselli, C. (2017)
Criminal network vulnerabilities and adaptations. Global Crime, 18,
424–441.

Bright, D., Koskinen, J. and Malm, A. (2019) Illicit network dynamics:
The formation and evolution of a drug trafficking network. Journal
of Quantitative Criminology, 35, 237–258.

Bright, D.A. (2015) Disrupting and dismantling dark networks: Lessons
from social network analysis and law enforcement simulations, in
Illuminating Dark Networks: The Study of Clandestine Groups and
Organizations., Cambridge University Press New York, NY, pp. 39–51.

Bright, D.A. and Delaney, J.J. (2013) Evolution of a drug trafficking
network: Mapping changes in network structure and function across
time. Global Crime, 14, 238–260.

Bright, D.A., Greenhill, C., Reynolds, M., Ritter, A. and Morselli, C.
(2015) The use of actor-level attributes and centrality measures to
identify key actors: A case study of an Australian drug trafficking
network. Journal of Contemporary Criminal Justice, 31, 262–278.

Bright, D.A., Hughes, C.E. and Chalmers, J. (2012) Illuminating dark
networks: A social network analysis of an Australian drug trafficking
syndicate. Crime, Law and Social Change, 57, 151–176.

Broccatelli, C., Everett, M. and Koskinen, J. (2016) Temporal
dynamics in covert networks. Methodological Innovations, 9,
205979911562276–205979911562214.

Brown, G.F. and Silverman, L.P. (1980) The retail price of heroin:
Estimation and applications, in Quantitative Explorations in Drug
Abuse Policy, Springer, Dordrecht, pp. 25–53.

Brown, S.S. and Hermann, M.G. (2020) Financing the illicit economy.
In Transnational Crime and Black Spots, International Political
Economy Series. Palgrave Macmillan, London, pp. 111–139.

Cai, P., Cai, J.-Y. and Naik, A.V. (1998) Efficient algorithms for a
scheduling problem and its applications to illicit drug market crack-
downs. Journal of Combinatorial Optimization, 1, 367–376.

Calderoni, F. (2012) The structure of drug trafficking mafias: The
Ndrangheta and cocaine. Crime, Law and Social Change, 58, 321–349.

Calderoni, F., Catanese, S., de Meo, P., Ficara, A. and Fiumara, G.
(2020) Robust link prediction in criminal networks: A case study of
the Sicilian mafia. Expert Systems with Applications, 161, 113666.

Calderoni, F. and Piccardi, C. (2014) Uncovering the structure of crim-
inal organizations by community analysis: The infinito network, in
2014 Tenth International Conference on Signal-Image Technology and
Internet-Based Systems. IEEE Press, Piscataway, NJ, pp. 301–308.

Camossi, E., Dimitrova, T. and Tsois, A. (2012) Detecting anomalous
maritime container itineraries for anti-fraud and supply chain secur-
ity, in 2012 European Intelligence and Security Informatics
Conference, IEEE Press, Piscataway, NJ, pp. 76–83.

Campana, P. and Varese, F. (in press) Studying organized crime net-
works: Data sources, boundaries and the limits of structural meas-
ures. Social Networks.

Carley, K.M. (2006) Destabilization of covert networks. Computational
and Mathematical Organization Theory, 12, 51–66.

Carley, K.M., Dombroski, M., Tsvetovat, M., Reminga, J. and
Kamneva, N. (2003) Destabilizing dynamic covert networks, in
Proceedings of the 8th International Command and Control Research
and Technology Symposium, pp. 79–92.

Carley, K.M., Lee, J.-S. and Krackhardt, D. (2002) Destabilizing net-
works. Connections, 24, 79–92.

20 R. ANZOOM ET AL.

Caulkins, J.P. (1993) Local drug markets’ response to focused police
enforcement. Operations Research, 41, 848–863.

Caulkins, J.P. (1995) Domestic geographic variation in illicit drug pri-
ces. Journal of Urban Economics, 37, 38–56.

Caulkins, J.P. (1997) Modeling the domestic distribution network for
illicit drugs. Management Science, 43, 1364–1371.

Caulkins, J.P. and Baker, D. (2010) Cobweb dynamics and price disper-
sion in illicit drug markets. Socio-Economic Planning Sciences, 44,
220–230.

Caulkins, J.P., Crawford, G. and Reuter, P. (1993) Simulation of adap-
tive response: A model of drug interdiction. Mathematical and
Computer Modelling, 17, 37–52.

Caulkins, J.P., Disley, E., Tzvetkova, M., Pardal, M., Shah, H. and
Zhang, X. (2016) Modeling the structure and operation of drug sup-
ply chains: The case of cocaine and heroin in Italy and Slovenia.
The International Journal on Drug Policy, 31, 64–73.

Caulkins, J.P. and Hao, H. (2008) Modelling drug market supply dis-
ruptions: Where do all the drugs not go? Journal of Policy Modeling,
30, 251–270.

Caulkins, J.P. and Padman, R. (1993) Interdiction’s impact on the
structure and behavior of the export-import sector for illicit drugs.
Zor – Methods and Models of Operations Research, 37, 207–224.

Cavallaro, L., Ficara, A., de Meo, P., Fiumara, G., Catanese, S.,
Bagdasar, O., Song, W. and Liotta, A. (2020) Disrupting resilient
criminal networks through data analysis: The case of Sicilian mafia.
Plos One, 15, e0236476.

Cedillo-Campos, M.G., S�anchez-Ram�ırez, C., Vadali, S., Villa, J.C. and
Menezes, M.B. (2014) Supply chain dynamics and the “cross-border
effect”: The US–Mexican border’s case. Computers & Industrial
Engineering, 72, 261–273.

Chaloupka, F.J., Edwards, S.M., Ross, H. and Diaz, M. (2015) Preventing
and reducing illicit tobacco trade in the United States. Technical
report, Centers for Disease Control and Prevention, Atlanta, GA.

Chandola, V., Banerjee, A. and Kumar, V. (2009) Anomaly detection:
A survey. ACM Computing Surveys, 41, 1–58.

Chen, Z., Van Khoa, L.D., Teoh, E.N., Nazir, A., Karuppiah, E.K. and
Lam, K.S. (2018) Machine learning techniques for anti-money laun-
dering solutions in suspicious transaction detection: A review.
Knowledge and Information Systems, 57, 245–285.

Cho, S.-H., Fang, X. and Tayur, S. (2015) Combating strategic counter-
feiters in licit and illicit supply chains. Manufacturing & Service
Operations Management, 17, 273–289.

Chopra, S. and Meindl, P. (2019) Supply Chain Management: Strategy,
Planning & Operation. Pearson Education.

Clemons, E.K., Reddi, S.P. and Row, M.C. (1993) The impact of infor-
mation technology on the organization of economic activity: The
“move to the middle” hypothesis. Journal of Management
Information Systems, 10, 9–35.

Clifton, K.L. and Rastogi, A. (2016) Curbing illegal wildlife trade: The
role of social network analysis. Technical Report 5, International
Union for the Conservation of Nature.

Coke-Hamilton, P. and Hardy, J. (2019, July) Illicit trade endangers the
environment, the law and the sdgs. We need a global response.
https://www.weforum.org/agenda/2019/07/illicit-trade-sdgs-environ-
ment-global-danger.

Coscia, M. and Rios, V. (2012) How and where do criminals operate?
Using Google to track Mexican drug trafficking organizations.
Technical report, Center for International Development at Harvard
University, Cambridge, MA.

Coutinho, J.A., Divi�ak, T., Bright, D. and Koskinen, J. (2020)
Multilevel determinants of collaboration between organised criminal
groups. Social Networks, 63, 56–69.

Crane, B.D. and Rivolo, A.R. (1997) An empirical examination of
counterdrug interdiction program effectiveness. Technical report,
Institute for Defense Analyses, Alexandria VA.

Crotty, S.M. (2015) Locating day-labor employment: Toward a geo-
graphic understanding of day-labor hiring site locations in the San
Diego metropolitan area. Urban Geography, 36, 993–1017.

Crotty, S.M. and Bouch�e, V. (2018) The red-light network: Exploring
the locational strategies of illicit massage businesses in Houston,
Texas. Papers in Applied Geography, 4, 205–227.

Das, K., Samanta, S. and Pal, M. (2018) Study on centrality measures in
social networks: A survey. Social Network Analysis and Mining, 8, 13.

Davis, R.C. and Lurigio, A.J. (1996) Fighting Back: Neighborhood
Antidrug Strategies. SAGE Publications, Thousand Oaks, CA.

Dean, G., Gottschalk, P. and Fahsing, I. (2010) Organized Crime:
Policing Illegal Business Entrepreneurialism. Oxford University Press.

Decker, S. and Chapman, M.T. (2008) Drug Smugglers on Drug
Smuggling: Lessons from the Inside. Temple University Press,
Philadelphia, PA.

Deville, D. (2013) The illicit supply chain, in Convergence: Illicit
Networks and National Security in the Age of Globalization.,
National Defence University Press, Washington DC, pp. 63–74.

Di Minin, E., Fink, C., Tenkanen, H. and Hiippala, T. (2018) Machine
learning for tracking illegal wildlife trade on social media. Nature
Ecology & Evolution, 2, 406–407.

Dicken, P. (2003) Global production networks in Europe and East
Asia: The automobile components industries. Technical report,
Manchester University, School of Environment and Development.
GPN Working Paper 7, Manchester, UK.

Diesner, J. and Carley, K.M. (2004) Using network text analysis to
detect the organizational structure of covert networks, in
Proceedings of the North American Association for Computational
Social and Organizational Science Conference.

Digiampietri, L.A., Roman, N.T., Meira, L.A., Ferreira, C.D., Kondo, A.A.,
Constantino, E.R., R.C. Brandao, B.C, R., Ribeiro, H.S. and Carolino,
P.K. (2008) Uses of artificial intelligence in the Brazilian customs fraud
detection system, in Proceedings of the 2008 International Conference on
Digital Government Research, pp. 181–187.

Dijkstra, E.W. (1959) A note on two problems in connexion with
graphs. Numerische Mathematik, 1, 269–271.

Dimitrov, N.B., Michalopoulos, D.P., Morton, D.P., Nehme, M.V., Pan,
F., Popova, E., Schneider, E.A. and Thoreson, G.G. (2011) Network
deployment of radiation detectors with physics-based detection prob-
ability calculations. Annals of Operations Research, 187, 207–228.

Dimitrova, T., Tsois, A. and Camossi, E. (2014) Development of a
web-based geographical information system for interactive visualiza-
tion and analysis of container itineraries. International Journal of
Computer and Information Technology, 3, 1–8.

Dittus, M., Wright, J. and Graham, M. (2018) Platform criminalism:
The ‘last-mile’ geography of the darknet market supply chain, in
Proceedings of the 2018 World Wide Web Conference, pp. 277–286.

Divi�ak, T., Dijkstra, J.K. and Snijders, T.A. (2019) Structure, multiplex-
ity, and centrality in a corruption network: The Czech Rath affair.
Trends in Organized Crime, 22, 274–297.

Dray, A., Mazerolle, L., Perez, P. and Ritter, A. (2008) Policing Australia’s
‘heroin drought’: Using an agent-based model to simulate alternative
outcomes. Journal of Experimental Criminology, 4, 267–287.

Duijn, P.A., Kashirin, V. and Sloot, P.M. (2014) The relative ineffect-
iveness of criminal network disruption. Scientific Reports, 4, 4238.

Elsisy, A., Mandviwalla, A., Szymanski, B. and Sharkey, T. (2020) A
synthetic network generator for covert network analytics. https://
arxiv.org/abs/2008.04445.

Enayaty-Ahangar, F., Rainwater, C.E. and Sharkey, T.C. (2019) A logic-
based decomposition approach for multi-period network interdiction
models. Omega, 87, 71–85.

Enders, W. and Su, X. (2007) Rational terrorists and optimal network
structure. Journal of Conflict Resolution, 51, 33–57.

Erd}os, P. and R�enyi, A. (1959) On random graphs I. Publicationes
Mathematicae Debrecen, 6, 290–297.

Everett, M.G. and Borgatti, S.P. (1999) The centrality of groups and
classes. The Journal of Mathematical Sociology, 23, 181–201.

Everton, S.F. (2009) Network topography, key players and terrorist net-
works. Connections, 32, 12–19.

Fanusie, Y. and Robinson, T. (2018) Bitcoin laundering: An analysis of
illicit flows into digital currency services. Technical report, Center
on Sanctions and Illicit Finance.

IISE TRANSACTIONS 21

https://www.weforum.org/agenda/2019/07/illicit-trade-sdgs-environment-global-danger

https://www.weforum.org/agenda/2019/07/illicit-trade-sdgs-environment-global-danger

https://arxiv.org/abs/2008.04445

https://arxiv.org/abs/2008.04445

Farasat, A., Gross, G., Nagi, R. and Nikolaev, A.G. (2016) Social net-
work analysis with data fusion. IEEE Transactions on Computational
Social Systems, 3, 88–99.

Farrugia, S., Ellul, J. and Azzopardi, G. (2020) Detection of illicit
accounts over the ethereum blockchain. Expert Systems with
Applications, 150, 113318.

Feige, E.L. (1997) Underground activity and institutional change:
Productive, protective and predatory behavior in transition econo-
mies, in Transforming Post-communist Political Economies, National
Academy Press, Washington DC, pp. 21–34.

Fortunato, S. (2010) Community detection in graphs. Physics Reports,
486, 75–174.

Fuentes, J.R. (1998) The life of a cell: Managerial practice and strategy
in Colombian cocaine distribution in the United States. Ph.D. thesis,
City University of New York.

Gaukler, G.M., Li, C., Ding, Y. and Chirayath, S.S. (2012) Detecting
nuclear materials smuggling: Performance evaluation of container
inspection policies. Risk Analysis: An Official Publication of the
Society for Risk Analysis, 32, 531–554.

Geller, A., Rizi, S.M.M. and ›atek, M.M. (2011) How corruption blunts
counternarcotic policies in Afghanistan: A multiagent investigation,
in International Conference on Social Computing, Behavioral-
Cultural Modeling, and Prediction, Springer, pp. 121–128.

Gera, R., Miller, R., MirandaLopez, M., Saxena, A. and Warnke, S.
(2017) Three is the answer: Combining relationships to analyze
multilayered terrorist networks, in 2017 IEEE/ACM International
Conference on Advances in Social Networks Analysis and Mining.,
IEEE Press, Piscataway, NJ, pp. 868–875.

Gimenez-Salinas Framis, A. (2011) Illegal networks or criminal organi-
zations: Power, roles and facilitators in four cocaine trafficking
structures, in Third Annual Illicit Networks Workshop, Montreal.

Giommoni, L., Aziani, A. and Berlusconi, G. (2017) How do illicit
drugs move across countries? A network analysis of the heroin sup-
ply to Europe. Journal of Drug Issues, 47, 217–240.

Godspower-Akpomiemie, E. and Ojah, K. (2019) Money laundering,
tax havens and transparency: Any role for the board of directors of
banks? in Enhancing Board Effectiveness, Routledge, New York, NY,
pp. 248–266.

Gonz�alez Ordiano, J. �A., Finn, L., Winterlich, A., Moloney, G. and
Simske, S. (2020a) A method for estimating driving factors of illicit
trade using node embeddings and clustering, Pattern Recognition,
Springer International Publishing, Cham, pp. 231–241.

Gonz�alez Ordiano, J. �A., Finn, L., Winterlich, A., Moloney, G. and
Simske, S. (2020b) On the analysis of illicit supply networks using
variable state resolution-Markov chains, in Information Processing
and Management of Uncertainty in Knowledge-Based Systems,
Springer International Publishing, Cham, pp. 513–527.

Grant Thornton (2013) Illicit trade in Ireland: Uncovering the cost to
the Irish economy. https://www.drugsandalcohol.ie/19844/1/Illicit-
Trade-in-Ireland-report .

Grassi, R., Calderoni, F., Bianchi, M. and Torriero, A. (2019)
Betweenness to assess leaders in criminal networks: New evidence
using the dual projection approach. Social Networks, 56, 23–32.

Greenhill, K.M. (2009) Kleptocratic interdependence: Trafficking, cor-
ruption, and the marriage of politics and illicit profits, in
Corruption, Global Security, and World Order, Brookings Institution
Press, Washington, DC, pp. 96–123.

Guo, Q., An, B., Zick, Y. and Miao, C. (2016) Optimal interdiction of
illegal network flow, in Proceedings of the Twenty-Fifth International
Joint Conference on Artificial Intelligence, pp. 2507–2513.

Haas, T.C. and Ferreira, S.M. (2015) Federated databases and actionable
intelligence: Using social network analysis to disrupt transnational wild-
life trafficking criminal networks. Security Informatics, 4, 2.

Hastings, J.V. (2012) The geography of nuclear proliferation networks:
The case of AQ Khan. The Nonproliferation Review, 19, 429–450.

Hauenstein, S., Kshatriya, M., Blanc, J., Dormann, C.F. and Beale, C.M.
(2019) African elephant poaching rates correlate with local poverty,
national corruption and global ivory price. Nature Communications,
10, 1–9.

Helbling, C.E., Kelly, C.D., Sipperly, C.D., Price, C.Z. and Schott, L.R.
(2012) Modeling Honduran illicit drug networks. Technical Report,
United States Military Academy, West Point, NY.

Hintsa, J. and Mohanty, S. (2014) A literature-based qualitative frame-
work for assessment of socio-economic negative impacts of common
illicit cross-border freight logistics flows, in Innovative Methods in
Logistics and Supply Chain Management. Proceedings of the
Hamburg International Conference of Logistics, Volume 18, epubli
GmbH, pp. 317–338.

Hirshman, J., Huang, Y. and Macke, S. (2013) Unsupervised
approaches to detecting anomalous behavior in the bitcoin transac-
tion network. Project for CS229, Stanford University, CA.

Hua, Z., Li, S. and Tao, Z. (2006) A rule-based risk decision-making
approach and its application in China’s customs inspection decision.
Journal of the Operational Research Society, 57, 1313–1322.

Hussain, D.A. and Ortiz-Arroyo, D. (2008) Locating key actors in
social networks using Bayes’ posterior probability framework, in
Intelligence and Security Informatics., Springer, Berlin Heidelberg,
pp. 27–38.

Isah, H., Neagu, D. and Trundle, P. (2015) Bipartite network model for
inferring hidden ties in crime data, in Proceedings of the 2015 IEEE/
ACM International Conference on Advances in Social Networks
Analysis and Mining, IEEE Press, Piscataway, NJ, pp. 994–1001.

Jabarzare, Z., Zolfagharinia, H. and Najafi, M. (2020) Dynamic inter-
diction networks with applications in illicit supply chains. Omega,
96, 102069.

Jacobson, S.H., Karnani, T., Kobza, J.E. and Ritchie, L. (2006) A cost-
benefit analysis of alternative device configurations for aviation-
checked baggage security screening. Risk Analysis: An Official
Publication of the Society for Risk Analysis, 26, 297–310.

Johns, T. and Hayes, R. (2003) Behind the fence: Buying and selling
stolen merchandise. Security Journal, 16, 29–44.

Kammer-Kerwick, M., Busch-Armendariz, N. and Talley, M. (2018)
Disrupting illicit supply networks: New applications of operations
research and data analytics to end modern slavery. Technical report,
Bureau of Business Research.

Kantor, P. and Boros, E. (2010) Deceptive detection methods for effect-
ive security with inadequate budgets: The testing power index. Risk
Analysis: An Official Publication of the Society for Risk Analysis, 30,
663–673.

Kilmer, B. and Hoorens, S. (2010) Understanding illicit drug markets,
supply-reduction efforts, and drug-related crime in the European
Union. Technical report, RAND Corporation.

Kinsella, D.T. (2008) The illicit arms trade: A social network analysis,
in ISA’s 49th Annual Convention, Bridging Multiple Divides. https://
pdxscholar.library.pdx.edu/polisci_fac/12/.

Kivel€a, M., Arenas, A., Barthelemy, M., Gleeson, J.P., Moreno, Y. and
Porter, M.A. (2014) Multilayer networks. Journal of Complex
Networks, 2, 203–271.

Kleiman, M. (2011) Surgical strikes in the drug wars: Smarter policies
for both sides of the border. Foreign Affairs, 90, 89–101.

Koen, H., de Villiers, J.P., Roodt, H. and de Waal, A. (2017) An
expert-driven causal model of the rhino poaching problem.
Ecological Modelling, 347, 29–39.

Konrad, R.A., Trapp, A.C., Palmbach, T.M. and Blom, J.S. (2017)
Overcoming human trafficking via operations research and analytics:
Opportunities for methods, models, and applications. European
Journal of Operational Research, 259, 733–745.

Kort, P.M., Feichtinger, G., Hartl, R.F. and Haunschmied, J.L. (1998)
Optimal enforcement policies (crackdowns) on an illicit drug mar-
ket. Optimal Control Applications and Methods, 19, 169–184.

Kovari, A. and Pruyt, E. (2012) Prostitution and human trafficking: A
model-based exploration and policy analysis, in Proceedings of the
30th International Conference of the System Dynamics Society.

Krebs, V.E. (2002) Mapping networks of terrorist cells. Connections,
24, 43–52.

Kretschmann, L. and M€unsterberg, T. (2017) Simulation-framework for
illicit-goods detection in large volume freight. In Digitalization in
Supply Chain Management and Logistics: Smart and Digital Solutions
for an Industry 4.0 Environment. Proceedings of the Hamburg

22 R. ANZOOM ET AL.

https://www.drugsandalcohol.ie/19844/1/Illicit-Trade-in-Ireland-report

https://www.drugsandalcohol.ie/19844/1/Illicit-Trade-in-Ireland-report

https://pdxscholar.library.pdx.edu/polisci_fac/12/

https://pdxscholar.library.pdx.edu/polisci_fac/12/

International Conference of Logistics, Vol. 23, epubli GmbH, Berlin,
pp. 427–448.

Kumar, R. and Tripathi, R. (2019) Traceability of counterfeit medicine
supply chain through blockchain, in 11th International Conference
on Communication Systems & Networks, IEEE Press, Piscataway, NJ,
pp. 568–570.

Lee, A.J., Nikolaev, A.G. and Jacobson, S.H. (2008) Protecting air trans-
portation: A survey of operations research applications to aviation
security. Journal of Transportation Security, 1, 160–184.

Levitt, S.D. and Venkatesh, S.A. (2000) An economic analysis of a
drug-selling gang’s finances. Quarterly Journal of Economics, 115,
755–789.

Lim, M., Abdullah, A. and Jhanjhi, N. (in press) Performance optimiza-
tion of criminal network hidden link prediction model with deep
reinforcement learning. Journal of King Saud University – Computer
and Information Sciences in Sciences.

Lindelauf, R., Borm, P. and Hamers, H. (2009) The influence of secrecy
on the communication structure of covert networks. Social
Networks, 31, 126–137.

Liu, K., Li, J.-A. and Lai, K.K. (2004) Single period, single product
newsvendor model with random supply shock. European Journal of
Operational Research, 158, 609–625.

Lozano, S. (2012) Information sharing in DEA: A cooperative game
theory approach. European Journal of Operational Research, 222,
558–565.

L€u, L. and Zhou, T. (2011) Link prediction in complex networks: A
survey. Physica A: Statistical Mechanics and Its Applications, 390,
1150–1170.

Mackey, T.K. and Kalyanam, J. (2017) Detection of illicit online sales
of fentanyls via twitter. F1000Research, 6, 1937.

Magliocca, N.R., McSweeney, K., Sesnie, S.E., Tellman, E., Devine, J.A.,
Nielsen, E.A., Pearson, Z. and Wrathall, D.J. (2019) Modeling
cocaine traffickers and counterdrug interdiction forces as a complex
adaptive system. Proceedings of the National Academy of Sciences of
the United States of America, 116, 7784–7792.

Malaviya, A., Rainwater, C. and Sharkey, T. (2012) Multi-period net-
work interdiction problems with applications to city-level drug
enforcement. IIE Transactions, 44, 368–380.

Malliaros, F.D. and Vazirgiannis, M. (2013) Clustering and community
detection in directed networks: A survey. Physics Reports, 533,
95–142.

Malm, A. and Bichler, G. (2011) Networks of collaborating criminals:
Assessing the structural vulnerability of drug markets. Journal of
Research in Crime and Delinquency, 48, 271–297.

Marciani, G., Porretta, M., Nardelli, M. and Italiano, G.F. (2017) A
data streaming approach to link mining in criminal networks, in 5th
International Conference on Future Internet of Things and Cloud
Workshops., IEEE Press, Piscataway, NJ, pp. 138–143.

Markowski, S., Koorey, S., Hall, P. and Brauer, J. (2009) Multi-channel
supply chain for illicit small arms. Defence and Peace Economics, 20,
171–191.

Martonosi, S.E., Ortiz, D.S. and Willis, H.H. (2007) Evaluating the via-
bility of 100 per cent container inspection at America’s ports. In
The Economic Impacts of Terrorist Attacks., Edward Elgar,
Northampton, MA, pp. 218–241.

Mashiri, E. and Sebele-Mpofu, F.Y. (2015) Illicit trade, economic
growth and the role of customs: A literature review. World Customs
Journal, 9, 38–50.

Mazerolle, L., Soole, D. and Rombouts, S. (2007) Drug law enforce-
ment: A review of the evaluation literature. Police Quarterly, 10,
115–153.

McCarthy, B., Hagan, J. and Cohen, L.E. (1998) Uncertainty, cooper-
ation, and crime: Understanding the decision to co-offend. Social
Forces, 77, 155–184.

McLay, L.A. and Dreiding, R. (2012) Multilevel, threshold-based poli-
cies for cargo container security screening systems. European
Journal of Operational Research, 220, 522–529.

McMasters, A.W. and Mustin, T.M. (1970) Optimal interdiction of a
supply network. Naval Research Logistics Quarterly, 17, 261–268.

Memon, B.R. (2012) Identifying important nodes in weighted covert
networks using generalized centrality measures, in 2012 European
Intelligence and Security Informatics Conference, IEEE Press,
Piscataway, NJ, pp. 131–140.

Meneghini, C., Aziani, A. and Dugato, M. (2020) Modeling the struc-
ture and dynamics of transnational illicit networks: An application
to cigarette trafficking. Applied Network Science, 5, 21.

Meng, N. (2013) Intelligent border patrol route optimization. Master’s
Thesis, Texas A&M University-Kingsville.

Michael, B. (2012) Do customs trade facilitation programs help reduce
customs-related corruption? International Journal of Public
Administration, 35, 81–97.

Miltenburg, J. (2018) Supply chains for iilicit products: Case study of
the global opiate production networks. Cogent Business &
Management, 5, 1423871.

Mitchel, A. (2005) The ESRI Guide to GIS Analysis, Volume 2: Spatial
Measurements and Statistics. Esri Press, Redlands, California.

Morselli, C. (2001) Structuring Mr. Nice: Entrepreneurial opportunities
and brokerage positioning in the cannabis trade. Crime, Law and
Social Change, 35, 203–244.

Morselli, C. (2009a) Hells Angels in springtime. Trends in Organized
Crime, 12, 145–158.

Morselli, C. (2009b) Inside Criminal Networks. Springer- Verlag New
York.

Morselli, C. (2010) Assessing vulnerable and strategic positions in a
criminal network. Journal of Contemporary Criminal Justice, 26,
382–392.

Morselli, C. and Petit, K. (2007) Law-enforcement disruption of a drug
importation network. Global Crime, 8, 109–130.

Morton, D.P., Pan, F. and Saeger, K.J. (2007) Models for nuclear smug-
gling interdiction. IIE Transactions, 39, 3–14.

Motoyama, M., McCoy, D., Levchenko, K., Savage, S. and Voelker,
G.M. (2011) An analysis of underground forums, in Proceedings of
the 2011 ACM SIGCOMM Conference on Internet Measurement, pp.
71–80.

Naik, A.V., Baveja, A., Batta, R. and Caulkins, J.P. (1996) Scheduling
crackdowns on illicit drug markets. European Journal of Operational
Research, 88, 231–250.

Natarajan, M. (2006) Understanding the structure of a large heroin dis-
tribution network: A quantitative analysis of qualitative data. Journal
of Quantitative Criminology, 22, 171–192.

Naylor, R.T. (2004) Wages of crime: Black markets, Illegal Finance, and
the Underworld Economy. Cornell University Press, Ithica, New
York.

Nehme, M.V. (2009) Two-person games for stochastic network inter-
diction: models, methods, and complexities. Ph.D. thesis, The
University of Texas at Austin.

Ozgul, F., Gok, M., Erdem, Z. and Ozal, Y. (2012) Detecting criminal
networks: SNA models are compared to proprietary models, in 2012
IEEE International Conference on Intelligence and Security
Informatics, IEEE Press, Piscataway, NJ, pp. 156–158.

Pan, F., Charlton, W.S. and Morton, D.P. (2003) A stochastic program
for interdicting smuggled nuclear material, in Network Interdiction
and Stochastic Integer Programming, Springer, Boston, MA, pp.
1–19.

Papachristos, A.V. and Smith, C. (2012) The small world of Al
Capone: The embedded and multiplex nature of organized crime.
http://dx.doi.org/10.2139/ssrn.2159899.

Parthasarathy, S., Ruan, Y. and Satuluri, V. (2011) Community discov-
ery in social networks: Applications, methods and emerging trends,
in Social Network Data Analytics., Springer, Boston, MA, pp.
79–113.

Patel, N.G., Rorres, C., Joly, D.O., Brownstein, J.S., Boston, R., Levy,
M.Z. and Smith, G. (2015) Quantitative methods of identifying the
key nodes in the illegal wildlife trade network. Proceedings of the
National Academy of Sciences of the United States of America, 112,
7948–7953.

Perera, S., Bell, M.G. and Bliemer, M.C. (2017) Network science
approach to modelling the topology and robustness of supply chain
networks: A review and perspective. Applied Network Science, 2, 33.

IISE TRANSACTIONS 23

http://dx.doi.org/10.2139/ssrn.2159899

Popping, R. (2000) Computer-assisted Text Analysis. SAGE
Publications, Thousand Oaks, CA.

Reuter, P. (2014) The mobility of drug trafficking. in Ending the Drug
Wars, Technical report, London, 33–40.

Robins, G. (2009) Understanding individual behaviors within covert
networks: The interplay of individual qualities, psychological predis-
positions, and network effects. Trends in Organized Crime, 12,
166–187.

Robinson, D. and Scogings, C. (2018) The detection of criminal groups
in real-world fused data: Using the graph-mining algorithm
“graphextract. Security Informatics, 7, 2.

Rydell, C.P., Caulkins, J.P. and Everingham, S.S. (1996) Enforcement or
treatment? Modeling the relative efficacy of alternatives for control-
ling cocaine. Operations Research, 44, 687–695.

Sadeghi, S. and Seifi, A. (2019) Stochastic maximum flow network
interdiction with endogenous uncertainty. International Journal of
Supply and Operations Management, 6, 200–212.

Sam, L.Z., bin Maarof, M.A., Selamat, A. and Shamsuddin, S.M. (2007)
Features extraction for illicit web pages identifications using inde-
pendent component analysis, in 2007 International Conference on
Intelligent and Advanced Systems., IEEE Press, Piscataway, NJ, pp.
139–144.

Sangkaran, T., Abdullah, A. and JhanJhi, N. (2020) Criminal network
community detection using graphical analytic methods: A survey.
EAI Endorsed Transactions on Energy Web, 7, e5.

Schneider, J.L. (2008) Reducing the illicit trade in endangered wildlife:
The market reduction approach. Journal of Contemporary Criminal
Justice, 24, 274–295.

Schroeder, J., Xu, J., Chen, H. and Chau, M. (2007) Automated crim-
inal link analysis based on domain knowledge. Journal of the
American Society for Information Science and Technology, 58,
842–855.

Schwartz, D.M. and Rouselle, T.D. (2009) Using social network analysis
to target criminal networks. Trends in Organized Crime, 12,
188–207.

Seddon, T. (2014) Drug policy and global regulatory capitalism: The
case of new psychoactive substances (NPS). International Journal of
Drug Policy, 25, 1019–1024.

Shaikh, M.A. and Jiaxin, W. (2008) Network structure mining:
Locating and isolating core members in covert terrorist networks.
WSEAS Transactions on Information Science and Applications, 5,
1011–1020.

Shan, X. and Zhuang, J. (2015) Subsidizing to disrupt a terrorism sup-
ply chain—a four-player game, in R.A. Forder, (ed), OR, Defence
and Security, The OR Essentials series. Palgrave Macmillan, London.

Sharkey, T.C., Cavdaroglu, B., Nguyen, H., Holman, J., Mitchell, J.E.
and Wallace, W.A. (2015) Interdependent network restoration: On
the value of information-sharing. European Journal of Operational
Research, 244, 309–321.

Shelley, L.I. (2018) Corruption & illicit trade. Daedalus, 147, 127–143.
Sherman, G., Siebers, P.-O., Menachof, D. and Aickelin, U. (2012)

Evaluating different cost-benefit analysis methods for port security
operations, in Decision Making in Service Industries: A Practical
Approach, CRC Press, Boca Raton, FL, pp. 279–303.

Sin, S. and Boyd, M. (2016) Searching for the nuclear silk road. In
Nuclear Terrorism: Countering the Threat., Routledge, New York,
NY, pp. 159–181.

Smith, J.C. and Song, Y. (2020) A survey of network interdiction mod-
els and algorithms. European Journal of Operational Research, 283,
797–811.

Soudijn, M. and Reuter, P. (2016) Cash and carry: The high cost of
currency smuggling in the drug trade. Crime, Law and Social
Change, 66, 271–290.

Staake, T., Thiesse, F. and Fleisch, E. (2009) The emergence of counter-
feit trade: A literature review. European Journal of Marketing, 43,
320–349.

Stevenson, R.J. and Forsythe, L. (1998) The stolen goods market
in New South Wales: An interview study with imprisoned bur-
glars. Technical report, NSW Bureau of Crime Statistics and
Research.

Suehr, S. and Vogiatzis, C. (2018) Now you see me: Identifying dupli-
cate network personas, in 2018 European Intelligence and Security
Informatics Conference, IEEE Press, Piscataway, NJ, pp. 23–30.

Taha, K. and Yoo, P.D. (2016) SIIMCO: A forensic investigation tool for
identifying the influential members of a criminal organization. IEEE
Transactions on Information Forensics and Security, 11, 811–822.

Taha, K. and Yoo, P.D. (2017) Using the spanning tree of a criminal
network for identifying its leaders. IEEE Transactions on
Information Forensics and Security, 12, 445–453.

Tammaro, A.M. (2014) Challenging illicit bulk cash flows: Next steps
for US law enforcement. Journal of Homeland Security and
Emergency Management, 11, 281–288.

Tezcan, B. and Maass, K.L. (2020) Human trafficking interdiction with
decision dependent success. engrxiv.org/dt8fs.

The Economist Intelligence Unit Limited (2018) The global illicit trade
environment index. http://illicittradeindex.eiu.com/.

Toledo, A., Carpi, L.C. and Atman, A. (2020) Diversity analysis exposes
unexpected key roles in multiplex crime networks, in Complex
Networks XI, Springer, Cham, pp. 371–382.

Tragler, G., Caulkins, J.P. and Feichtinger, G. (2001) Optimal dynamic
allocation of treatment and enforcement in illicit drug control.
Operations Research, 49, 352–362.

Transnational Alliance to Combat Illicit Trade (2019) Mapping the impact of
illicit trade on the sustainable development goals. https://unctad.org/meet-
ings/en/Contribution/DITC2020/_TRACIT/_IllicitTradeandSDGs/_fullre-
port/_en .

Triepels, R., Daniels, H. and Feelders, A. (2018) Data-driven fraud
detection in international shipping. Expert Systems with
Applications, 99, 193–202.

Tsirogiannis, C. and Tsirogiannis, C. (2016) Uncovering the missing
routes: An algorithmic study on the illicit antiquities trade network,
in Across Space and Time., Amsterdam University Press,
Amsterdam, The Netherlands, pp. 508–515.

Tsvetovat, M. and Carley, K.M. (2003) Bouncing back: Recovery mech-
anisms of covert networks, in NAACSOS Conference, Pittsburgh, PA,
Carnegie-Mellon University.

Turner, J. and Kelly, L. (2009) TRADE SECRETS: Intersections
between diasporas and crime groups in the constitution of the
human trafficking chain. British Journal of Criminology, 49,
184–201.

Van Der Veen, H. (2003) Taxing the drug trade: Coercive exploitation
and the financing of rule. Crime, Law and Social Change, 40,
349–390.

Wagner, R.A. and Fischer, M.J. (1974) The string-to-string correction
problem. Journal of the ACM, 21, 168–173.

Washburn, A. and Wood, K. (1995) Two-person zero-sum games for
network interdiction. Operations Research, 43, 243–251.

Wasserman, S. and Faust, K. (1994) Social Network Analysis: Methods
and Applications, Cambridge University Press, Cambridge, UK.

Watson, P. and Todeschini, C. (2007) The Medici Conspiracy: The
Illicit Journey of Looted Antiquities–From Italy’s Tomb Raiders to the
World’s Greatest Museum. PublicAffairs, New York, NY.

Watts, D.J. and Strogatz, S.H. (1998) Collective dynamics of ’small-
world’ networks. Nature, 393, 440–442.

WCO (2017) Illicit trade report 2017.
Wein, L.M., Liu, Y., Cao, Z. and Flynn, S.E. (2007) The optimal spatio-

temporal deployment of radiation portal monitors can improve
nuclear detection at overseas ports. Science & Global Security, 15,
211–233.

Williams, K.C. (2020) Parallel imports and the principle of exhaustion:
The first sale rule in international commerce. Journal of Law and
International Affairs Blog. https://sites.psu.edu/jlia/parallel-imports-
and-the-principle-of-exhaustion-the-first-sale-rule-in-international-
commerce/.

Williams, P. and Godson, R. (2002) Anticipating organized and
transnational crime. Crime, Law and Social Change, 37,
311–355.

Williamson, O.E. (1975) Markets and Hierarchies. Free Press, New
York.

24 R. ANZOOM ET AL.

http://illicittradeindex.eiu.com/

https://unctad.org/meetings/en/Contribution/DITC2020/_TRACIT/_IllicitTradeandSDGs/_fullreport/_en

https://unctad.org/meetings/en/Contribution/DITC2020/_TRACIT/_IllicitTradeandSDGs/_fullreport/_en

https://unctad.org/meetings/en/Contribution/DITC2020/_TRACIT/_IllicitTradeandSDGs/_fullreport/_en

Parallel Imports and the Principle of Exhaustion: The First Sale Rule in International Commerce

Parallel Imports and the Principle of Exhaustion: The First Sale Rule in International Commerce

Parallel Imports and the Principle of Exhaustion: The First Sale Rule in International Commerce

Wilt, J. and Sharkey, T.C. (2019) Measuring the impact of coordination
in disrupting illicit trafficking networks, in IISE Annual Conference
Proceedings, pp. 767–772.

Wollmer, R. (1964) Removing arcs from a network. Operations
Research, 12, 934–940.

Wood, R.K. (1993) Deterministic network interdiction. Mathematical
and Computer Modelling, 17, 1–18.

World Economic Forum (2012) Network of global agenda councils
reports 2011-2012. https://reports.weforum.org/global-agenda-coun-
cil-2012/#cover.

Xu, J. and Chen, H. (2008) The topology of dark networks.
Communications of the ACM, 51, 58–65.

Xu, J.J. and Chen, H. (2004) Fighting organized crimes: Using shortest-
path algorithms to identify associations in criminal networks.
Decision Support Systems, 38, 473–487.

Xu, J.J. and Chen, H. (2005) Crimenet explorer: A framework for crim-
inal network knowledge discovery. ACM Transactions on
Information Systems, 23, 201–226.

Yaqin, W. and Yuming, S. (2010) Classification model based on associ-
ation rules in customs risk management application, in 2010
International Conference on Intelligent System Design and
Engineering Application, IEEE Press, Piscataway, NJ, pp. 436–439.

Yen, J.Y. (1970) An algorithm for finding shortest routes from all
source nodes to a given destination in general networks. Quarterly
of Applied Mathematics, 27, 526–530.

Zhang, J., Zhuang, J. and Behlendorf, B. (2018) Stochastic shortest path
network interdiction with a case study of Arizona–Mexico border.
Reliability Engineering & System Safety, 179, 62–73.

Zhao, F., Skums, P., Zelikovsky, A., Sevigny, E.L., Swahn, M.H.,
Strasser, S.M., Huang, Y. and Wu, Y. (2020) Computational
approaches to detect illicit drug ads and find vendor communities
within social media platforms. IEEE/ACM Transactions on
Computational Biology and Bioinformatics Early Access, 1–1.

Zhao, M. (2019) The illicit distribution of precursor chemicals in
China: A qualitative and quantitative analysis. International Journal
for Crime, Justice and Social Democracy, 8, 106–120.

IISE TRANSACTIONS 25

https://reports.weforum.org/global-agenda-council-2012/#cover

https://reports.weforum.org/global-agenda-council-2012/#cover

• Abstract

Introduction
Review methods and statistics
Illicit trade

Special Issue: Analytical Methods for Detecting, Disrupting, and Dismantling Illicit Operations

IISE Transactions: Focused Issue on Operations Engineering and Analytics

We highly encourage authors to submit abstracts to the lead editor (tcshark@clemson.edu) by October 31, 2021,
so the editorial team to provide feedback on the submission and to facilitate a timely review of the full paper.

Illicit operations, which operate outside of the boundaries of law, are a significant global issue, with some
estimates valuing their economic value as upwards of $1.6 trillion each year. Categories of illicit operations
include, but are not limited to: drug trafficking; human trafficking; illegal mining, logging, and fishing; wildlife
trafficking; money laundering; supply chains producing counterfeit goods; organ trafficking; identity theft; and
weapons trafficking. This special issue will highlight analytical approaches that can help detect, disrupt, and
ultimately dismantle illicit operations. Its goal is to showcase the role of analytical methods in the fight against
illicit operations by bringing together those within the industrial and systems engineering (ISE) community and
domain experts in distinct areas of fighting illicit operations. We are especially interested in papers that integrate
domain expertise and/or stakeholder engagement in their model formulation, analysis, and validation.

Each type of illicit operation has its own nuances, which should be carefully considered as analytical approaches
are created to help detect, disrupt, and dismantle the operation. These nuances motivate the need for methods that
integrate research across disciplines to create analytical approaches to help the fight against illicit operations,
along with broad activities for this fight, including socio-economic approaches or interventions from non-profits
that decrease vulnerabilities and/or demand. Further, because illicit operations are hidden, data to populate
analytical methods may be difficult to gather. This special issue encourages papers that (i) pull in perspectives
from multiple disciplines in demonstrating the impact of the approach on a specific type of illicit operation, (ii)
explore broad approaches to detect, disrupt, and dismantle illicit operations, and/or (iii) address issues around data
on illicit operations and its role in analytical approaches. Submissions bringing together ISE and expertise from
disciplines such as anthropology, criminology, economics, geography, public policy, and sociology are
encouraged. Topics covered include, but are not limited to:

• Models to help understand the behavior of
specific types of illicit operations

• Predictive methods to understand how illicit
operations evolve and adapt over time

• Models that help understand the interactions of
multiple types of illicit operations

• Models and methods that can examine how policy
shapes illicit markets

• Analytical methods to identify, gather, and
analyze data on illicit operations

• Analytical frameworks to structure human
judgement in illicit contexts

• Prescriptive methods to identify effective
strategies that disrupt illicit operations, as
opposed to simply displacing them

• Models which account for a socio-economic
perspective reflective of the environment giving
rise to the illicit operation

• Models and methods that examine and reduce
the demand patterns for illicit goods

• Methods to evaluate the role of culture and
environment in facilitating illicit operations

Papers must be submitted through http://mc.manuscriptcentral.com/iietransactions and prepared according to the
journal’s Instructions for authors. Select “Special Issue” for the question “Please select the Focus Issue to which
the paper is most related” at Step 1 in the submission process, and select the specific special issue at Step 6.

Important Dates
• (Encouraged) Abstract Submission: 10/31/2021 • Completion of 2nd round review: 9/30/2022
• Manuscript submission: 1/31/2022 • Final manuscript submission: 11/30/2022
• Completion of 1st round review: 4/30/2022 • Tentative publication date: 12/2022

Guest Editors Focus Issue Editor
Dr. Thomas Sharkey Dr. Renata Konrad Dr. J. Cole Smith
Clemson University Worcester Polytechnic Institute Syracuse University
tcshark@clemson.edu rkonrad@wpi.edu colesmit@syr.edu

Dr. Burcu B. Keskin Dr. Maria E. Mayorga
University of Alabama North Carolina State University
bkeskin@cba.ua.edu memayorg@ncsu.edu

Alvari et al. Secur Inform (2017) 6:1
DOI 10.1186/s13388-017-0029-8

R E S E A R C H

Semi-supervised learning for detecting
human trafficking
Hamidreza Alvari1* , Paulo Shakarian1 and J. E. Kelly Snyder

2

Abstract
Human trafficking is one of the most atrocious crimes and among the challenging problems facing law enforcement
which demands attention of global magnitude. In this study, we leverage textual data from the website “Backpage”—
used for classified advertisement—to discern potential patterns of human trafficking activities which manifest online
and identify advertisements of high interest to law enforcement. Due to the lack of ground truth, we rely on a human
analyst from law enforcement, for hand-labeling a small portion of the crawled data. We extend the existing Laplacian
SVM and present S3VM − R, by adding a regularization term to exploit exogenous information embedded in our fea-
ture space in favor of the task at hand. We train the proposed method using labeled and unlabeled data and evaluate
it on a fraction of the unlabeled data, herein referred to as unseen data, with our expert’s further verification. Results
from comparisons between our method and other semi-supervised and supervised approaches on the labeled data
demonstrate that our learner is effective in identifying advertisements of high interest to law enforcement.

Keywords: Human trafficking, Backpage, Semi-supervised support vector machines, Laplacian support vector
machines

© The Author(s) 2017. This article is distributed under the terms of the Creative Commons Attribution 4.0 International License
(http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium,
provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license,
and indicate if changes were made.

Background
According to the United Nation  [1], human trafficking
is defined as the modern slavery or the trade of humans
mostly for the purpose of sexual exploitation and forced
labor, via different improper ways including force, fraud
and deception. The United States’ Trafficking Victim
Protection Act of 2000 (TVPA 2000) [2] was the first US
legislation passed against human trafficking. Human traf-
ficking has ever since received increased national and
societal concern  [3] but still demands persistent fight
against from all over the globe. No country is immune
and the problem is rapidly growing with little to no
law enforcement addressing the issue. This problem is
amongst the challenging ones facing law enforcement as
it is difficult to identify victims and counter traffickers.

Before the advent of the Internet, human traffickers
were under risks of being arrested by law enforcement
while advertising their victims on streets  [4]. How-
ever, move to the Internet has made it easier and less

dangerous for sex sellers  [5] as they no longer needed
to advertise on the streets. There are now a plethora of
websites that host and provide sexual services under cat-
egories of escort, adult entertainment, massage services,
etc., which help sex sellers and buyers maintain their
anonymity. Although some services such as the Craig-
list’s adult section and myredbook.com were shut down
recently, there are still many websites such as the Back-
page.com that provide such services and many new are
frequently created. Traffickers even use dating and social
networking websites such as the Twitter, Facebook, Ins-
tagram and Tinder to reach out to sex buyers and their
followers. Although the Internet has presented new traf-
ficking related challenges for law enforcement, it has
also provided readily and publicly available rich source
of information which could be gleaned from online sex
advertisements for fighting this crime  [6]. However,
the problem is we lack the ground truth and obtaining
the labels through hand-labeling is indeed tedious and
expensive even for a small subset of data—this is the
point where the semi-supervised setting comes in handy.

Despite considerable attention which has been
devoted to studying supervised, unsupervised and

Open Access

*Correspondence: halvari@asu.edu
1 Arizona State University, Tempe, Arizona, USA
Full list of author information is available at the end of the article

http://orcid.org/0000-0001-7172-0134

http://creativecommons.org/licenses/by/4.0/

http://crossmark.crossref.org/dialog/?doi=10.1186/s13388-017-0029-8&domain=pdf

Page 2 of 14Alvari et al. Secur Inform (2017) 6:1

semi-supervised learning settings via different applica-
tions  [7–13], semi-supervised learning, i.e., learning
from labeled and unlabeled examples, is still one of the
most interesting yet challenging problems in the machine
learning community [14]. The idea is simple though—we
shall have an approach that makes a better use of unla-
beled data to boost performance. This is pretty close to
the most natural learning that occurs in the world. For
the most part, we as humans are exposed only to a small
number of labeled instances; yet we successfully general-
ize well by effective utilization of a large amount of unla-
beled data. This motivates us to use unlabeled samples
to improve recognition performance while developing
classifiers.

In this article, expanding on our previous work  [15],
we use the data crawled from the adult entertainment
section of the website Backpage.com and extend the
existing Laplacian SVM framework [14] to detect escort
advertisements of high interest to law enforcement.
Here, we merely focus on the online advertisements
although the Internet has triggered many other activi-
ties including attracting the victims, communicating
with customers and rating the escort services. We thus
highlight several contributions of the current research as
follows.

1. Based on the literature, we created different groups
of features that capture the characteristics of poten-
tial human trafficking activities. The less likely
human trafficking related posts were then filtered
out using these features. We also conducted a fea-
ture importance analysis to demonstrate how these
features contribute to the proposed learner.

2. We extended the Laplacian SVM [14] and proposed
the semi-supervised support vector machine learn-
ing algorithm, S3VM − R. In particular, we incor-
porated additional information of our feature space
as a regularization term into the standard optimiza-
tion formulation with regard to the Laplacian SVM.
We also used geometry of the underlying data as an
intrinsic regularization term in Laplacian SVM.

3. We trained our model on both of the labeled and
unlabeled data and sent back the identified human
trafficking related advertisements to an expert from
law enforcement for further verification. We then
validated our approach on a small subset of the unla-
beled data (i.e. unseen data) with further verification
of the expert.

4. We performed comparisons between our approach
and several semi-supervised and supervised base-
lines on both of the labeled and unseen data (so-
called blind evaluation).

5. We demonstrated the effect of varying different
hyperparameters used in our learner on its perfor-
mance.

The rest of the paper is organized as follows. In section
“Related work”, we review the prior studies on human
trafficking. Section “Data preparation” covers our data
preparation, feature engineering, unsupervised filtering
and expert assisted labeling. We detail our semi-super-
vised learning approach in section “Semi-supervised
learning framework” by deriving the required equations.
Section “Experimental study” provides in-depth explana-
tion of our experiments. Section “Conclusion” concludes
the paper by providing future research directions.

Related work
Recently, several studies have examined the role of the
Internet and related technology in facilitating human
trafficking [16–18]. For example, the work of [16] studied
how closely sex trafficking is intertwined with new tech-
nologies. According to [17], sexual exploitation of women
and children is a global human right crisis that is being
escalated by the use of new technologies. Researchers
have studied relationships between new technologies and
human trafficking and advantages of the Internet for sex
traffickers. For instance, findings from a group of experts
from the Council of Europe demonstrated that the Inter-
net and sex industry are closely interlinked and volume
and content of the material on the Internet promoting
human trafficking are unprecedented [18].

One of the earliest works which leveraged data min-
ing techniques for online human trafficking was  [18],
wherein the authors conducted data analysis on the adult
section of the website Backpage.com. Their findings con-
firmed that female escort post frequency would increase
in Dallas, Texas, leading up to the Super Bowl 2011 event.
In a similar attempt, other studies  [19, 20] have investi-
gated impact of large public events such as the Super
Bowl on sex trafficking by exploring advertisement vol-
ume, trends and movement of advertisements along with
the scope and volume of demand associated with such
events. The work of  [19] for instance, concluded that
large events such as the Super Bowl which attract sig-
nificant amount of concentration of people in a relatively
short period of time and in a confined urban area, could
be a desirable location for sex traffickers to bring their
victims for commercial sexual exploitation. Similarly, the
data-driven approach of [20] showed that in some but not
all events, one can see a correlation between occurrence
of the event and statistically significant evidence of an
influx of sex trafficking activity. Also, certain studies [21]
have tried to build large distributed systems to store and

Page 3 of 14Alvari et al. Secur Inform (2017) 6:1

process available online human trafficking data in order
to perform entity resolution and create ontological rela-
tions between entities.

Beyond these works, the work of  [22] studied the
problem of isolating sources of human trafficking from
online advertisements with a pairwise entity resolution
approach. Specifically, they used phone number as a
strong feature and trained a classifier to predict if two ads
are from the same source. This classifier was then used
to perform entity resolution using a heuristically learned
value for the score of classifier. Another work of [6] used
Backpage.com data and extracted most likely human
trafficking spatio-temporal patterns with the help of law
enforcement. Note that unlike our method, this work
did not employ any machine learning methodologies
for automatically identifying human trafficking related
advertisements. The work of [23] also deployed machine
learning for the advertisement classification problem, by
training a supervised learning classifier on labeled data
(based on phone numbers of known traffickers) provided
by a victim advocacy group. We note that while phone
numbers can provide a very precise set of positive labeled
data, there are clearly many posts with previously unseen
phone numbers.

In contrast, we do not solely rely on phone numbers for
labeling our data. Instead, our expert analyze each post’s
content to identify whether it is human trafficking related
or not. To do so, we first filter out most likely advertise-
ments using several feature groups and pass a small sam-
ple to the expert for hand-labeling. Then, we train our
semi-supervised learner on both of the labeled and unla-
beled data which in turn lets us evaluate our approach
on new coming (unseen) data later. We note that our
semi-supervised approach can also be used as a comple-
mentary method to procedures such as those described
in [23] as we can significantly expand the training set for
use with supervised learning.

Finally, note that our current research is different from
our previous work [15] and we list the key nuances here:

• In this study we experiment with a much larger data-
set. To obtain such dataset, we use the same raw data
from  [15], but this time with slight modifications of
the thresholds that were used for filtering out less
likely human trafficking related advertisements.

• As opposed to our previous research which deployed
only one feature space, in this work, two feature
spaces that have complementary roles to each other
are used.

• In this paper we present a new framework based on
the existing Laplacian SVM  [14], by adding a regu-
larization term to the standard optimization problem
and solving the new optimization equation derived

from there. In contrast, [15] utilized the off-the-shelf
graph based semi-supervised learner, LabelSpreading
method [24], without any further manipulation of the
original approach.

• Unlike [15] in which we did not compare our method
with other approaches, this work compares our pro-
posed framework against other semi-supervised and
supervised learners. Also unlike our previous work
in which only one group of human trafficking related
advertisements were passed to two experts for valida-
tion, here in order to reduce the inconsistency, two
control groups of advertisements–those of interest to
law enforcement and those of not—are sent to only
one expert for verification.

Data preparation
We collected about 20K publicly available listings from the
US posted on Backpage.com in March, 2016. Each post
includes a title, description, time stamp, poster’s age, post-
er’s ID, location, image, and sometimes video and audio.
The description usually lists the attributes of the
individual(s) and contact phone numbers. In this work, we
only focus on the textual component of the data. This free-
text data required significant cleaning due to a variety of
issues common to textual analytics (i.e. misspellings, for-
mat of phone numbers, etc.). We also acknowledge that the
information in data could be intentionally inaccurate, such
as poster’s name, age and even physical appearance (e.g. bra
cup size, weight). Figure 1 shows an actual post from Back-
page.com. To illustrate geographic diversity of the listings,
we use the Tableau1 software to visualize choropleth map of
phone frequency with respect to the different states in
Fig. 2, wherein darker colors mean higher frequencies.

Next, we will explain most important characteristics
of potential human trafficking advertisements which are
captured by our feature groups.

Feature engineering
Though many advertisements on Backpage.com are
posted by posters selling their own services without
coercion and intervention of traffickers, some do exhibit
many common trafficking triggers. For example, in con-
trast to Fig.  1, Fig.  3 shows an advertisement that could
be an evidence of human trafficking. This advertisement
indicates several potential properties of human traffick-
ing, including advertising for multiple escorts with the
first individual coming from Asia and very young. In
what follows, such common properties of human traf-
ficking related advertisements are discussed in more
detail.

1 https://www.tableau.com/.

https://www.tableau.com/

Page 4 of 14Alvari et al. Secur Inform (2017) 6:1

Inspired by the literature, we define and extract 6
groups of features from advertisements (see Table  1).
These features could be amongst the strong indicators
of human trafficking. Let us now briefly describe each
group of features used in our work. Note each feature
listed here is ultimately treated as a binary variable.

Advertisement language pattern
The first group consists of different language related
features. For the first and second features, we identify
posts which have third person language (more likely to

be written by someone other than the escort) and posts
which contain first person plural pronouns such as ‘we’
and ‘our’ (more likely to be an organization) [6].

To ensure their anonymity, traffickers would deploy
techniques to generate diverse information and hence
make their posts look more complicated. They usually
do this to avoid being identified by either human analysts
or automated programs. Thus, to obtain the third feature
we take an approach from complexity theory, namely
Kolmogorov complexity, which is defined as length of
shortest program to reproduce a string of characters on
a universal machine such as the Turing Machine  [25].
Since the Kolmogorov complexity is not computable, we
approximate the complexity of an advertisement con-
tent by first removing stop words and then computing
entropy of the content [25]. To illustrate this, let X denote
the content and xi be a given word in the content. We use
the following equation [31] to calculate the entropy of the
content and thus approximate the Kolmogorov complex-
ity of X:

We expect higher values of the entropy correspond
to human trafficking. Finally, we discretize the result by
using the threshold of 4 which was found empirically in
our experiments.

(1)K(X) ≈

−

n

∑

i=

1

P(xi) log2 P(xi

)

Fig. 1 A real post from Backpage.com. To ensure anonymity, the
personal information has been intentionally obfuscated

Fig. 2 Choropleth map of phone frequency with respect to the different states. Darker colors show higher frequencies

Page 5 of 14Alvari et al. Secur Inform (2017) 6:1

For the next features, we use word-level n-grams to
find common language patterns of advertisements. This
particular choice is because of the fact that character-
level n-grams have already shown to be useful in detect-
ing unwanted content for spam detection  [26]. We set
n = 4 and use the range of (4,4) to compute normalized
n-grams (using TF-IDF) of each advertisement content.
We ultimately create a matrix whose rows and columns
correspond to the advertisements contents and their
associated 4-grams, respectively. We rank all elements of
this matrix in a descending order and pick the top 3 ones.
Finally for each advertisement content, 3 elements with
the column numbers associated with the top elements
are chosen. This way, 3 more features will be added to our
feature set. Overall, we have 6 features related to the lan-
guage of the advertisement.

Words and phrases of interest
Despite the fact that advertisements on Backpage.com
do not directly mention sex with children, customers
who prefer children know to look for words and phrases
such as “sweet, candy, fresh, new in town, new to the
game”  [27–29]. We thus investigate within the posts to
see if they contain such words as they could be highly
related with human trafficking in general.

Countries of interest
We identify if the individual being escorted is coming
from other countries such as those in Southeast Asia
(especially from China, Vietnam, Korea and Thailand, as
we observed in our data) [3].

Multiple victims advertised
Some advertisements advertise for multiple women at
the same time. We consider the presence of more than
one victim as a potential evidence of organized human
trafficking [6].

Victim weight
We take into account the weight of the individual being
escorted as a feature (if it is available). This information is
particularly useful assuming that for the most part, lower
body weights (under 115 lbs) correlate with smaller and
underage girls [2, 30] and thereby human trafficking.

Reference to website or spa massage therapy
The presence of a link in the advertisement either refer-
encing to an outside website (especially infamous ones) or
spa massage therapy could be an indicator of more elab-
orate organization  [6]. In particular, in case of spa ther-
apy, we observed many advertisements interrelated with
advertising for young Asian girls and their erotic mas-
sage abilities. Therefore, the last group of features has two
binary features for presence of any website and spa.

Finally, in order to extract all of the above features, we
first clean the original data and conduct preprocessing. By
applying these features, we draw a random sample of 3543
instances out of our dataset for further analysis to see if
they are evidences of human trafficking—this is described
in the next section.

Unsupervised filtering
Having detailed our feature set, we now construct a fea-
ture vector for each instance by creating a vector of 12
binary features that correspond to the important charac-
teristics of human trafficking. Hereafter, we refer to this
feature space, as our first feature space and denote it with
F1. As mentioned earlier, we draw 3543 instances from
our raw data by filtering out those that do not posses any
of the binary features. We will refer to this as our filtered

Fig. 3 An evidence of human trafficking. The boxes and numbers in
red, indicate the features and their corresponding group numbers
(see also Table 1)

Table 1 Different features and their corresponding groups

No. Feature group References

1 Advertisement language pattern [6, 25, 26]

Third person language

First person plural pronouns

Kolmogorov complexity

n-grams (1)

n-grams (2)

n-grams (3)

2 Words and phrases of interest [27–29]

3 Countries of interest [3]

4 Multiple victims advertised [6]

5 Victim weight [2, 30]

6 Reference to website or spa massage therapy [6]

Reference to a website

Reference to a spa massage therapy

Page 6 of 14Alvari et al. Secur Inform (2017) 6:1

dataset. For the sake of visualization, a 2-D projection
(using the t-SNE transformation  [32]) of the filtered
dataset is depicted in Fig. 4. The purpose of this figure is
to demonstrate how hard it is for basic clustering tech-
niques such as the K-means, to correctly assign labels to
unlabeled instances using only few existing labeled ones.

Now, we shall define our second feature space, namely
F2, which will be used to compute geometry of the
underlying data. Note that our proposed framework will
utilize both of the feature spaces in the form of regulari-
zation terms, to detect advertisements of high interest
to law enforcement. After conducting standard preproc-
essing techniques on the filtered dataset, we build F2 by
transforming the filtered data into a 3543 ×  3543 matrix
of TF-IDF similarity features. Each entry in this matrix
simply shows the similarity between a pair of advertise-
ments in our filtered dataset.

Note that since we lack the ground truth, we would rely
on a human analyst (expert) for labeling the listings as
either ‘of interest’ or ‘of not interest’ to law enforcement.
In the next section, we select a smaller yet finer grain
subset of this data to be sent to the expert. This alleviates
the burden of the tedious work of hand-labeling.

Expert assisted labeling
We first obtain a sample of 200 listings from the filtered
dataset. This set of listings was labeled by our expert from
law enforcement who is specialized in this type of crime.
From this subset, the law enforcement professional

identified 70 instances to be of interest to law enforce-
ment and the rest to be not human trafficking related.
However, we are still left with a large amount of the unla-
beled examples (3343 instances) in our dataset. The ratio
of the labeled to unlabeled instances in our dataset is very
small (about 0.06). The statistics of our dataset is summa-
rized in Table 2.

Semi‑supervised learning framework
Here, we first introduce some preliminary notations
necessary for the rest of the discussion and then outline
our proposed semi-supervised approach, S3VM − R, for
detecting online human trafficking. Note as said earlier,
our framework is an extension to the existing Laplacian
SVM  [14]. In particular, we incorporated another regu-
larization term into the standard Laplacian SVM to lever-
age the additional information of our first feature space
and then solved the associated optimization problem.
Consequently, similar notation is adopted throughout the

Fig. 4 2-D projection of the entire set of the filtered data

Table 2 Description of the dataset

Name Value

Raw 20,822

Filtered 3543

Unlabeled 3343

Labeled Positive Negative

70 130

Page 7 of 14Alvari et al. Secur Inform (2017) 6:1

following section. Furthermore, we shall once again note
that our current research does not utilize any off-the-
shelf graph based semi-supervised leaner in contrast to
our previous research [15].

Technical preliminaries
We assume a set of l labeled pairs {(xi, yi)}li=1 and an
unlabeled set of u instances {xl+i}ui=1, where xi ∈ R

n and
yi ∈ {+1, −1}. Recall for the standard soft-margin sup-
port vector machine, the following optimization problem
is solved:

In the above equation, fθ(·) is a decision function of the
form fθ(·) = w.�(·) + b where θ = (w, b) are the param-
eters of the model, and �(·) is the feature map which is
usually implemented using the kernel trick  [33]. Also,
the function H1(·) = max(0, 1 − ·) is the Hinge Loss
function.

The classical Representer theorem  [34] suggests that
solution to the optimization problem exists in a Hilbert
space Hk and is of the following form:

where K is the l × l Gram matrix over labeled samples.
Equivalently, the above problem can be written as:

Next, we will use the above optimization equation as
our basis to derive the formulations for our proposed
semi-supervised learner.

The proposed method
The basic assumption behind semi-supervised learning
methods is to leverage unlabeled instances in order to
restructure hypotheses during the learning process. In
this paper, exogenous information extracted from both of
our feature spaces is further exploited to make a better
use of the unlabeled examples. To do so, we first intro-
duce matrix F in F1 and over both of the labeled and
unlabeled samples with Fij defined as follows:

(2)

min
fθ ∈Hk

γ ||fθ||
2

k + Cl

l
∑

i=1

H1(yifθ(xi))

(3)f

∗

θ (x) =

l
∑
i=1

α

∗

i K(x, xi)

(4)min
w,b,ǫ

1

2
||w||22 + Cl

l
∑
i=1

ǫi

(5)
s.t. yi(w.�(xi) + b)

≥ 1 − ǫi, i = 1, . . . , l

ǫi ≥ 0, i = 1, . . . , l

(6)Fij =
1

nf
(�(xi) · �(xj))

where nf is the number of features in F1 (here, nf = 12 ).
We force the instances xi and xj in our dataset to have
same label if they both possess same features. To account
for this, a regularization term is added to the standard
equation and the following optimization is solved:

where f = [f (x1), . . . , f (xl+u)]T and L is the Lapla-
cian matrix based on F given by L = D − F, and
Dii =

∑l+u
j=1 Fij. The intuition here is that any two

instances which are composed of same features are more
likely to have same labels than others. Next, by solving a
similar optimization problem, we are able to capture data
geometry in F2 as fTθ L

′T fθ (also referred to as the intrin-
sic smoothness penalty term [14]). Here, L′ is the Lapla-
cian of matrix A associated with the data adjacency graph
G in F2.

We construct G with (l + u) nodes in F2, and by add-
ing an edge between each pair of nodes 〈i, j〉, if the edge
weight Wij exceeds a given threshold. For computing the
edge weights, we use the heat kernel [35] as a function of
the Euclidean distance between two samples in F2, hence
we set Wij = exp−||xi−xj||

2/4t.
Following the notations used in  [14] and by including

our regularization term as well as the intrinsic smooth-
ness penalty term, we would extend the standard equa-
tion by solving the following optimization:

Note one typical value for the smoothness penalty coef-
ficient Cs is

γI

(l+u)2

, where 1
(l+u)2

is a natural scale factor for
empirical estimate of the Laplace operator and γI is a reg-
ularization term  [14]. Again, solution in Hk would be in
the following form:

Here K is the (l + u) × (l + u) Gram matrix over all sam-
ples. The Eq. 8 could be then written as follows:

(7)min
fθ ∈Hk

1
2
l
∑
i=1

Fij||fθ(xi) − fθ(xj)||
2
2 = f

T
θ L

T

fθ

(8)

min
fθ ∈Hk

γ ||fθ ||
2

k + Cl
l
∑
i=1

H1(yifθ (xi)) + Crf
T
θ Lfθ + Csf

T
θ L

′

fθ

(9)f
∗

θ (x) =

l+u
∑

i=1
α
∗
i K(x, xi)

(10)

min
α,b,ǫ

1

2
α
T

Kα

+ Cl

l
∑
i=1

ǫi

+

Cr

2
α
T
KLKα

+
γI

2(l + u)2
α
T
KL

′
Kα

Page 8 of 14Alvari et al. Secur Inform (2017) 6:1

With introduction of the Lagrangian multipliers β and
γ, we write the Lagrangian function of the above equa-
tion as follows:

Obtaining the dual representation, requires taking the
following steps:

With the above equations, we formulate the reduced
Lagrangian as a function of only α and β as follows:

This equation is further simplified as follows:

(11)

s.t. yi





l+u

�

j=1

αjK(xi, xj) + b





≥ 1 − ǫi, i = 1, . . . , l
ǫi ≥ 0, i = 1, . . . , l

(12)

L(α, ǫ, b, β, γ ) =
1

2
α
T

K

�

I +

CrL +
γI

(l + u)2
L
′

�

α + Cl

l
�

i=1
ǫi
−
l
�
i=1

βi



yi




l+u
�

j=1
αjK(xi, xj) + b



− 1 + ǫi



−
l
�
i=1

γiǫi

(13)
∂L

∂b
= 0 →

l
∑
i=1

βiyi = 0

(14)
∂L

∂ǫi
= 0 → Cl − βi − γi = 0 →

0 ≤ βi ≤ Cl

(15)

LR(α, β) =
1

2
α
T
K
�
I + CrL +
γI
(l + u)2
L
′
�
α
−
l
�
i=1
βi

yi


l+u
�
j=1
αjK(xi, xj) + b


− 1 + ǫi



(16)

L
R
(α, β) =

1
2
α
T
K

(

I + CrL +
γI
(l + u)2
L
′
)
α

− α
T
KJ

T
Yβ +

l
∑
i=1
βi

In the above equation, J = [I 0] is a l × (l + u) matrix, I is
the l × l identity matrix and Y is a diagonal matrix con-
sisting of the labels of the labeled examples.

In the followings, we first take the derivative of LR with
respect to α and then set ∂L

R(α,β)
∂α

= 0:

Accordingly, we obtain α∗ by solving the following
equation:

Next, we obtain the dual problem in the form of a
quadratic programming problem by substituting α back
in the reduced Lagrangian function:

where β = [β1, . . . , βl]T ∈ Rl are the Lagrangian multi-
pliers and Q is obtained as follows:

We summarize the proposed semi-supervised frame-
work in Algorithm  1. Our optimization problem is very
similar to the standard optimization problem solved for
SVMs, hence we use a standard optimizer for SVMs to
solve our problem.

(17)K
(

I + CrL +
γI
(l + u)2
L
′
)

α − K

J
T
Y

β = 0

(18)α∗ =
(

I + CrL +
γI
(l + u)2
L
′
)

−1

J
T
Yβ

∗

(19)β
∗

= argmaxβ∈Rl −
1

2
β
T
Qβ +

l
∑
i=1
βi

(20)
s.t.

l
∑
i=1
βiyi = 0
0 ≤ βi ≤ Cl

(21)Q = YJK
(

I +

(
CrL +
γI
(l + u)2
L
′
)
K
)
−1
J
T
Y

Page 9 of 14Alvari et al. Secur Inform (2017) 6:1

Experimental study
In this section, we provide a comprehensive analysis of
the proposed framework by designing a series of experi-
ments on the filtered dataset. First, we explain several
approaches used in this study. Next, various results are
discussed: (1) comparisons on the labeled data were
made between our method and other approaches, (2)
experiments were performed on a fraction of the unla-
beled data (i.e., unseen data), and the results were further
verified by our expert to see what fraction is of interest
to law enforcement, (3) blind evaluation was conducted
to examine other approaches on the unseen data, and
finally, (4) experiments were designed to analyze effect
of varying different hyperparameters on our method as
well as impact of different groups of features in F1 on our
approach.

Approaches
We present results for the following methods:

• Semi-supervised S3VM − R, Laplacian support
vector machines  [14], graph inference based label
spreading approach  [24] with radial basis function
(RBF) and K-nearest neighbors (KNN) kernels, and
co-training learner  [36] with two support vector
machines classifiers (SVM).

• Supervised SVM, KNN, Gaussian naïve Bayes, logis-
tic regression, adaboost and random forest.

For the sake of fair comparison, all algorithms were
implemented and run in Python. More specifically, the
Python package CVXOPT2 was used to implement
S3VM − R and Laplacian support vector machines, and
all other approaches were implemented with the help of
the Scikit-learn3 package in Python. Note for those meth-
ods that require special tuning of parameters, we per-
formed grid search to choose the best set of parameters.
Before going any further, we first define main parameters
used in each method and then demonstrate their best
values picked by our grid search. The discussion on the
effect of varying the hyperparameters on our learner is
provided in the section “Hyperparameter sensitivity”.

• S3VM − R we set the penalty parameter as Cl = 0.6
and the regularization parameters Cr = 0.2 and
Cs = 0.2. Linear kernel was used in our approach.

• Laplacian SVM we used linear kernel and set the
parameters Cl = 0.6 and Cs = 0.6.

2 http://cvxopt.org/.
3 http://scikit-learn.org/stable/.

• LabelSpreading (RBF) RBF Kernel was used and γ
was set to the default value of 20.

• LabelSpreading (KNN) KNN kernel was used and the
number of neighbors was set to 5.

• Co-training (SVM) we followed the algorithm intro-
duced in  [36] and used two SVM as our classifiers.
For both SVMs we set the tolerance for stopping cri-
teria to 0.001 and the penalty parameter C = 1.

• SVM tolerance for stopping criteria was set to the
default value of 0.001. Penalty parameter C was set to
1 and linear kernel was used.

• KNN number of neighbors was set to 5.
• Gaussian NB there were no specific parameter to

tune.
• Logistic regression we used the ‘l2’ penalty. We also

set the parameter C = 1 (the inverse of regularization
strength) and tolerance for stopping criteria to 0.01.

• Adaboost number of estimators was set to 200 and
we also set the learning rate to 0.01.

• Random forest we used 200 estimators and the
‘entropy’ criterion was used.

Classification results
Here, we first evaluate the entire set of approaches on a
small portion of the data for which we already know the
labels, i.e., the labeled examples. We note that expert-
generated judgmental labeling might be error-prone,
though it is served as a surrogate to the ground truth
problem.

We used tenfold cross-validation on the labeled data in
the following way. We first divided the set of the labeled
samples into 10 different sets of approximately equal size.
Each time we held one set out for validation (by removing
their labels and adding them to the unlabeled samples)
and used the remaining along with the unlabeled samples
for the training–this was performed for all approaches
for the sake of fair comparison. Finally, we reported the
average of 10 different runs, using different combina-
tions of the feature spaces and various evaluation met-
rics, including the area under curve (AUC), accuracy,
precision, recall and F1-score. In Table  3, we reported
the average AUC and accuracy for each method and each
feature space. On the other hand, for precision, recall
and F1-score, we reported separate results for each fea-
ture space, in Tables 4, 5 and 6, respectively. Note, each of
these tables includes separate scores for the positive and
negative classes. In general, we observed the followings:

• Overall, our approach achieved highest performance
on F1 (Tables  3, 4) and {F1, F2} (Table  6), in terms
of all metrics. However it did not perform well using
solely F2 (Table  5), i.e. when Cr = 0. This clearly
demonstrates the importance of using Cr over Cs.

http://cvxopt.org/

http://scikit-learn.org/stable/

Page 10 of 14Alvari et al. Secur Inform (2017) 6:1

Table 3 AUC and accuracy results with tenfold cross‑validation on the labeled data

The best performance is in italic

Learner AUC Accuracy

F1 F2 {F1, F2}

F1 F2 {F1, F2}

S
3
VM − R 0.91 0.9 0.96 0.91 0.9 0.97

Laplacian SVM 0.9 0.9 0.9 0.91 0.9 0.92

LabelSpreading (RBF) 0.78 0.87 0.84 0.8 0.85 0.86

LabelSpreading (KNN) 0.68 0.80 0.74 0.71 0.8 0.8

Co-training (SVM) 0.82 0.94 0.92 0.85 0.94 0.93

SVM 0.82 0.9 0.91 0.85 0.92 0.93

KNN 0.76 0.91 0.81 0.79 0.92 0.84

Gaussian NB 0.78 0.91 0.9 0.82 0.9 0.9

Logistic regression 0.82 0.89 0.88 0.85 0.92 0.92

AdaBoost 0.82 0.85 0.85 0.85 0.88 0.88

Random forest 0.81 0.89 0.89 0.83 0.91 0.92

Table 4 Precision, recall and F1‑score for the positive and negative classes using

F1

Experiments were run using tenfold cross-validation on the labeled data. The best performance is in italic

Learner Precision Recall F1-score

classp classn classp classn classp classn

S
3
VM − R 0.91 0.92 0.91 0.93 0.91 0.92

Laplacian SVM 0.86 0.89 0.88 0.9 0.87 0.88

LabelSpreading (RBF) 0.76 0.78 0.77 0.73 0.8 0.81

LabelSpreading (KNN) 0.65 0.7 0.71 0.68 0.69 0.73

Co-training (SVM) 0.81 0.84 0.71 0.92 0.73 0.87

SVM 0.86 0.83 0.68 0.96 0.74 0.88

KNN 0.72 0.8 0.63 0.88 0.65 0.83

Gaussian NB 0.79 0.81 0.72 0.85 0.73 0.81

Logistic regression 0.81 0.85 0.71 0.93 0.74 0.88

AdaBoost 0.86 0.83 0.68 0.95 0.74 0.88

Random forest 0.77 0.85 0.73 0.89 0.73 0.86

Table 5 Precision, recall and F1‑score for the positive and negative classes using F2

Experiments were run using tenfold cross-validation on the labeled data. The best performance is in italic
Learner Precision Recall F1-score
classp classn classp classn classp classn

S
3
VM − R 0.91 0.9 0.9 0.9 0.91 0.92

Laplacian SVM 0.91 0.9 0.9 0.91 0.89 0.92

LabelSpreading (RBF) 0.8 0.86 0.82 0.83 0.81 0.85

LabelSpreading (KNN) 0.7 0.75 0.73 0.78 0.79 0.77

Co-training (SVM) 0.96 0.91 0.91 0.97 0.93 0.93

SVM 0.93 0.91 0.84 0.97 0.87 0.93

KNN 0.87 0.92 0.88 0.94 0.87 0.93

Gaussian NB 0.78 0.96 0.94 0.87 0.84 0.91

Logistic regression 0.98 0.89 0.81 0.98 0.88 0.93

AdaBoost 0.88 0.88 0.75 0.95 0.78 0.91

Random forest 0.93 0.89 0.81 0.97 0.85 0.93

Page 11 of 14Alvari et al. Secur Inform (2017) 6:1

• When the feature space used is F2, Co-training
(SVM) is the best method. Next best methods are
supervised learners KNN and Gaussian NB. Three
remarks can be made here. First, our approach
could not always defeat supervised learners as it is
seen from Tables  3 and  5. This is not surprising and
in fact lies at the inherent difference between semi-
supervised and supervised methods—unlabeled
examples could make the trained model susceptible
to error propagation and thus wrong estimation. Sec-
ond, as it is seen in Tables 4, 5 and 6, achieving very
high recall on the negative examples and low score
on the positive ones shall not be treated as a potent
property, otherwise a trivial classifier which always
assigns negative labels to all samples would be the
best learner. Third, using Cr always improves the per-
formance over Cs. One point that needs to be clari-
fied is, our ultimate goal is not to achieve high per-
formance on the labeled data, but rather to detect the
suspicious (unlabeled) advertisements which could
be human trafficking related—this will be explained
in more details in “Blind evaluation”.

• Compared to the other semi-supervised approaches,
our approach either achieved higher or comparable
AUC scores. The reason we performed exactly the
same as the Laplacian SVM, is because by setting
Cr = 0, the two approaches are inherently the same.

• For the Laplacian SVM to be able to run on F1, the
Laplacian L′ has to be constructed using F1 while
inherently is supposed to be made using F2. This
is because Cr is essentially associated with F1, and
Cs corresponds to L′ and correspondingly F2. The
same holds for {F1, F2}, where we need to construct
a new feature space by concatenating F1 and F2 as

the Laplacian SVM does not inherently use F1 at all.
The new feature space is then used to construct the
Laplacian L′.

• Since our approach inherently incorporates both of
the Laplacian matrices corresponding to the two fea-
ture spaces F1 and F2, all other baselines were also
run using the concatenation of these two feature
spaces for the sake of fair comparison. Unlike our
approach which used the wise combination of F1 and
F2, other methods do not gain high AUC by simply
combining the feature spaces.

Blind evaluation
For the next set of experiments, we first run our method
on the entire filtered dataset and without cross-vali-
dation. Recall from the previous sections that this is to
make a better use of the unlabeled examples. Then the
following control experiment was conducted. Our learner
was tested on the whole set of the unlabeled examples.
Out of 3343 instances, our approach identified two sets
of positive and negative instances. The positive set con-
tained 394 advertisements which were likely to be of
interest to law enforcement, whereas the negative set
included the remaining 2962 unlabeled advertisements
of probably less interest to law enforcement. Next, to
precisely determine the correctly identified fractions of
these two sets, we randomly picked two subsets (control
groups) of 100 examples from each set for further valida-
tion by our expert.

We passed these two control groups to our expert for
further verification. The expert-validated results demon-
strated that all of the examples in the positive group were

Table 6 Precision, recall and F1‑score for the positive and negative classes using {F1, F2}

Experiments were run using tenfold cross-validation on the labeled data. The best performance is in italic
Learner Precision Recall F1-score
classp classn classp classn classp classn

S
3
VM − R 0.97 0.97 0.95 0.98 0.94 0.95

Laplacian SVM 0.96 0.94 0.91 0.96 0.91 0.93

LabelSpreading (RBF) 0.83 0.86 0.82 0.84 0.81 0.86

LabelSpreading (KNN) 0.71 0.74 0.75 0.78 0.8 0.78

Co-training (SVM) 0.92 0.9 0.9 0.94 0.91 0.92

SVM 0.96 0.92 0.84 0.97 0.89 0.94

KNN 0.84 0.83 0.67 0.95 0.73 0.88

Gaussian NB 0.77 0.96 0.94 0.87 0.84 0.91

Logistic regression 0.95 0.9 0.79 0.97 0.85 0.93

AdaBoost 0.88 0.88 0.75 0.95 0.78 0.91

Random forest 0.93 0.9 0.82 0.97 0.86 0.93

Page 12 of 14Alvari et al. Secur Inform (2017) 6:1

of interest to law enforcement, while only two examples
from the negative group were not correctly classified as
of not being of any interest to law enforcement. Thus,
both results support the effectiveness of our framework
in identifying highly human trafficking advertisements.
Using the same two control groups and AUC metric, we
now perform so-called blind evaluation (see Table  7) of
other baselines. Note, we call this blind since actual labels
are not provided and the expert-generated labels might
convey uninformative information. In general, supervised
methods failed to achieve good results in the blind evalu-
ation compared to most of the semi-supervised methods.

Hyperparameter sensitivity
Here, we discuss how altering the hyperparameters Cl, Cr
and Cs may affect the performance of S3VM − R. We
start off by fixing the value of Cl to 0.6, which was empiri-
cally found to work well in our experiments. Also, recall
from the previous sections that one typical choice for Cs
is γI

(l+u)2
 [14]. Here, we set Cs = 0.2 and varied the values

of Cr as {0, 0.0002, 0.0006, 0.2, 1.0} and plotted the results

in Fig.  5. We used the same tenfold cross-validation set-
ting from the previous section.

We made the following observation. With the slight
increase of Cr, the performance of our approach
increased, peaked and then stabilized, i.e., further
increase of Cr did not change the performance. This sug-
gests significance of deploying the additional information
from our first feature space F1, over F2 and its corre-
sponding smoothness penalty parameter Cs which is used
by S3VM − R and the standard Laplacian SVM.

Next, to see the impact of Cl on the performance, we
set Cr = 0.2 and varied Cl as {0.2, 0.4, 0.6, 0.8, 1.0}. The
results are depicted in Fig. 5. We note that setting Cl = 0
is meaningless and thus we do not have any performance
corresponding to that—otherwise each βi in Eq. 19 would
be zero. In general, the performance was not particular
sensitive to this parameter—varying by 0.2 for values of
0.4 and greater.

Finally, having fixed Cl = 0.6 and Cr = 0.2, we also
tried other values for Cs including

∑l+u
i,j=1 Wij suggested

by [14] and depicted the results in Fig. 5. The results sug-
gest that our approach is less sensitive to this parameter
compared to Cr and Cl.

Significance of features
To examine how much discriminative our feature groups
in F1 are, we further conducted an analysis using the
labeled examples and the standard feature selection
measure χ2 to find the top features—only half of the
features with scores greater than a given threshold (0.5)
were selected (see Table 8 for the complete set of features
and their corresponding χ2 scores).

From this list, we noticed that ‘countries of interest’ and
‘reference to spa massage therapy’ were the most discrim-
inative feature groups, while ‘advertisement language
pattern’ group (with 3 important features) appeared to be
the most dominant feature group.

Figure  6 compares the top features against the less
important subset of the features (denoted by F∗1 ) in the
filtered dataset, in terms of frequency values. Note for
clarity, we have removed from this figure, the features
with frequency less than 20. According to this figure, our

Table 7 Blind evaluation of  the baselines on  the two con‑
trol groups

The best performance is in italic

Learner AUC

F1 F2 {F1, F2}

Laplacian SVM 0.9 0.92 0.93

LabelSpreading (RBF) 0.75 0.85 0.87

LabelSpreading (KNN) 0.7 0.82 0.79

Co-training (SVM) 0.8 0.9 0.91

SVM 0.8 0.65 0.69

KNN 0.74 0.62 0.77

Gaussian NB 0.77 0.51 0.52

Logistic regression 0.76 0.62 0.75

AdaBoost 0.77 0.74 0.74

Random forest 0.8 0.8 0.8

Fig. 5 Effect of varying different parameters on the performance

Page 13 of 14Alvari et al. Secur Inform (2017) 6:1

most discriminative features are not necessarily those
that appear more often.

To further investigate the importance of each of the top
features, we performed classification using the labeled
examples and the previous setting, on basis of these two
subsets of the features and their combination, i.e., F∗1, F

∗

1
and F1. The classification results are shown in Table  9.
We made the following observations:

• Considering only the feature space F1, our approach
achieved higher performance compared to all other
baselines by either using the whole feature space or
the most discriminative features F∗1.

• Deploying only the features from F∗1, we were able to
achieve comparable results as if we used the whole
feature space F1.

Conclusion
Readily available online data from escort advertisements
could be leveraged in favor of fight against human traf-
ficking. In this study, having focused on textual informa-
tion from the available data crawled from Backpage.com,
we identified if an escort advertisement can be reflective
of human trafficking activities. In particular, we first pro-
posed an unsupervised filtering approach to filter out
the data which are more likely involved in human traf-
ficking. We then proposed a semi-supervised learner,
namely S3VM − R, and trained it on a small portion of
the data which was hand-labeled by a human trafficking
expert. We used the trained model to identify labels of
unseen data. Results suggested our approach is effec-
tive at identifying potential human trafficking related
advertisements.

Our future plans include replicating the study by
integrating more interesting features especially those
supported by the criminology literature. Also, since
hand-labeling unlabeled examples is expensive, an inter-
esting research direction would be to deploy active
learning to enable iterative supervised learning to
actively query the user for labels. We also note that real-
world data is often more imbalanced compared to our
data, and the reason is that number of negative samples
usually outweigh positive ones. We would thus like to
apply the proposed framework on a more realistic data-
set which contains much less suspicious posts than nor-
mal posts.

Authors’ contributions
HA developed and implemented the human trafficking detection approach
and drafted the manuscript. PS provided guidance through the whole
project and revised the manuscript. JS was in contact with an expert from law
enforcement who was responsible for hand-labeling portions of the data. All
authors read and approved the final manuscript.

Author details
1 Arizona State University, Tempe, Arizona, USA. 2 Find Me Group, Tempe,
Arizona, USA.

Acknowledgements
This work was funded by the Find Me Group, a 501(c)3 dedicated to bring
resolution and closure to families of missing persons. The authors would like
to thank anonymous reviewers for their valuable suggestions to improve the
quality of the paper.

Competing interests
The authors declare that they have no competing interests.

Table 8 Significance of the features in F1

The check-marked features show the top features

No. Feature group χ2 Selected

1 Advertisement language pattern

Third person language 8.4 �

First person plural pronouns 9.5 �

Kolmogorov complexity 0.7 �

n-grams (1) 0.4

n-grams (2) 0.0

n-grams (3) 0.4

2 Words and phrases of interest 0.0

3 Countries of interest 59.3 �

4 Multiple victims advertised 14.1 �

5 Victim weight 0.2

6 Reference to website or spa massage therapy

Reference to website 0.1

Reference to spa massage therapy 33.5 �

Fig. 6 Frequency of each feature in F1 in the filtered dataset. Fea-
tures are grouped into the two groups, most important (F∗

1
) and less

important features (F∗
1

), according to χ2

Table 9 Classification results (AUC) using tenfold cross‑
validation and  different subsets of  the features on  the
labeled data

Name Value

F
∗
1

F
∗
1
F1

S
3
VM − R 0.82 0.87 0.91

Page 14 of 14Alvari et al. Secur Inform (2017) 6:1

Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in pub-
lished maps and institutional affiliations.

Received: 20 December 2016 Accepted: 9 April 2017

References
1. (2011) UNODC on human tracking and migrant smuggling. https://www.

unodc.org/unodc/en/human-tracking/. Accessed 20 Dec 2016
2. (2000) Trafficking victims protection act of 2000. https://www.state.gov/j/

tip/laws/61124.htm. Accessed 20 Dec 2016
3. (2015) Trafficking in persons report. https://www.state.gov/j/tip/rls/

tiprpt/2015/. Accessed 20 Dec 2016
4. Desplaces C (1992) Police run ‘Prostitution’ sting; 19 men arrested,

charged in Fourth East Dallas operation. Dallas Morning News
5. Kristof ND (2012) How pimps use the web to sell girls. New York Times.

http://www.nytimes.com/2012/01/26/opinion/how-pimps-use-the-web-
to-sell-girls.html. Accessed 20 Dec 2016

6. Kennedy E (2012) Predictive patterns of sex trafficking online. Dietrich
College Honors Theses, B.S. thesis, Carnegie Mellon University

7. Mitchell TM (2006) Learning from labeled and unlabeled data. Mach
Learn 10:701

8. Beigi G, Tang J, Liu H (2016) Signed link analysis in social media networks.
In: 10th international conference on web and social media, ICWSM 2016,
AAAI Press, Cologne, Germany

9. Backstrom L, Leskovec J (2011) Supervised random walks: predict-
ing and recommending links in social networks. In: Proceedings of
the fourth ACM international conference on Web search and data
mining. WSDM ’11, Hong Kong, China. ACM, New York, pp 635–644.
doi:10.1145/1935826.1935914

10. Alvari H, Hajibagheri A, Sukthankar G, Lakkaraju K (2016) Identifying
community structures in dynamic networks. Soc Netw Anal Min SNAM
6(1):77. doi:10.1007/s13278-016-0390-5

11. Beigi G, Tang J, Wang S, Liu H (2016) Exploiting emotional information for
trust/distrust prediction. In: Proceedings of the 2016 SIAM international
conference on data mining (ICDM), SIAM, Miami, FL, USA

12. Mitchell et al TM (1997) Machine learning, I–XVII. McGraw-Hill Education.
https://www.bibsonomy.org/bibtex/2e1eee0d0daaef20092093d6643b53
c4f/machinelearning. Accessed 20 Dec 2016

13. Beigi G, Jalili M, Alvari H, Sukthankar G (2014) Leveraging community
detection for accurate trust prediction. In: ASE international conference
on social computing, Palo Alto, CA

14. Belkin M, Niyogi P, Sindhwani V (2006) Manifold regularization: a geomet-
ric framework for learning from labeled and unlabeled examples. J Mach
Learn Res 7(Nov):2399–2434

15. Alvari H, Shakarian P, Snyder J (2016) A non-parametric learning approach
to identify online human trafficking. In: IEEE intelligence and security
informatics (ISI) conference, pp. 133–138

16. Hughes DM et al (2005) The demand for victims of sex trafficking. Wom-
ens Studies Program, University of Rhode Island, RI

17. Hughes DM (2002) The use of new communications and information
technologies for sexual exploitation of women and children. Hastings
Women’s Law J 13(1):129–148

18. Latonero M (2011) Human trafficking online: the role of social networking
sites and online classifieds. Available at SSRN 2045851

19. Roe-Sepowitz D, Gallagher J, Bracy K, Cantelme L, Bayless A, Larkin J,
Reese A, Allbee L (2015) Exploring the impact of the super bowl on sex
trafficking. https://www.mccaininstitute.org/exploring-the-impact-of-
the-super-bowl-on-sex-tracking-2015/. Accessed 20 Dec 2016

20. Miller K, Kennedy E, Dubrawski A (2016) Do public events affect sex traf-
ficking activity? https://arxiv.org/abs/1602.05048. Accessed 20 Dec 2016

21. Szekely PA, Knoblock CA, Slepicka J, Philpot A, Singh A, Yin C, Kapoor D,
Natarajan P, Marcu D, Knight K, Stallard D, Karunamoorthy SS, Bojanapalli
R, Minton S, Amanatullah B, Hughes T, Tamayo M, Flynt D, Artiss R, Chang
S-F, Chen T, Hiebel G, Ferreira L (2015) Building and using a knowledge
graph to combat human trafficking. In: International semantic web
conference (2), ser. Lecture notes in computer science, vol 9367. Springer,
Switzerland, pp 205–221

22. Nagpal C, Miller K, Boecking B, Dubrawski A (2015) An entity resolution
approach to isolate instances of human trafficking online. https://arxiv.
org/abs/1509.06659. Accessed 20 Dec 2016

23. Dubrawski A, Miller K, Barnes M, Boecking B, Kennedy E (2015) Leveraging
publicly available data to discern patterns of human-trafficking activity. J
Hum Traffick 1(1):65–85. doi:10.1080/23322705.2015.1015342

24. Zhou D, Bousquet O, Lal TN, Weston J, Schlkopf B (2004) Learning with
local and global consistency. In: Advances in neural information process-
ing systems 16. MIT Press, Cambridge, MA, pp. 321–328

25. Li M, Vitnyi PM (2008) An introduction to kolmogorov complexity and its
applications, 3rd edn. Springer, New York

26. Kanaris I, Kanaris K, Stamatatos E (2006) Spam detection using character
n-grams. In: Antoniou G, Potamias G, Spyropoulos C, Plexousakis D (eds),
Advances in Artificial Intelligence. SETN 2006. Lecture Notes in Computer
Science, vol 3955. Springer, Berlin, pp 95–104

27. Hetter K (2012) Fighting sex trafficking in hotels, one room at a time.
Cable News Network. http://www.cnn.com/2012/02/29/travel/hotel-sex-
tracking/. Accessed 20 Dec 2016

28. Lloyd R (2012) An open letter to jim Buckmaster. Change.org Inc.
http://www.hungtonpost.com/rachel-lloyd/an-open-letter-to-jim-
buc_b_570666.html. Accessed 20 Dec 2016

29. Dickinson Goodman J, Holmes M (2011) Can we use RSS to catch rapists.
Grace Hopper Celebration of Women in Computing, Portland, OR. http://
jessicadickinsongoodman.com/2011/10/30/can-we-use-rss-to-catch-
rapists-poster-nished/. Accessed 20 Dec 2016

30. Average height to weight chart—babies to teenagers. http://www.disa-
bled-world.com/artman/publish/height-weight-teens.shtml. Accessed 20
Dec 2016

31. Shannon CE (2001) A mathematical theory of communication. ACM
SIGMOBILE Mob Comput Commun Rev 5(1):3–55

32. van der Maaten L, Hinton G (2008) Visualizing high-dimensional data
using t-SNE. Journal of Machine Learning Research 9:2579–2605

33. Cortes C, Vapnik V (1995) Support-vector networks. Mach Learn
20(3):273–297

34. Belkin M, Niyogi P, Sindhwani V (2004) On manifold regularization. Techni-
cal report tr-2004-05. The University of Chicago

35. Grigoryan A (2006) Heat kernels on weighted manifolds and applications.
Contempl Math 398:93–191

36. Blum A, Mitchell T (1998) Combining labeled and unlabeled data with
co-training. In: Proceedings of the eleventh annual conference on Com-
putational learning theory. ACM, pp 92–100

https://www.unodc.org/unodc/en/human-tracking/

https://www.unodc.org/unodc/en/human-tracking/

https://www.state.gov/j/tip/laws/61124.htm

https://www.state.gov/j/tip/laws/61124.htm

https://www.state.gov/j/tip/rls/tiprpt/2015/

https://www.state.gov/j/tip/rls/tiprpt/2015/

http://dx.doi.org/10.1145/1935826.1935914

http://dx.doi.org/10.1007/s13278-016-0390-5

https://www.bibsonomy.org/bibtex/2e1eee0d0daaef20092093d6643b53c4f/machinelearning

https://www.bibsonomy.org/bibtex/2e1eee0d0daaef20092093d6643b53c4f/machinelearning

https://www.mccaininstitute.org/exploring-the-impact-of-the-super-bowl-on-sex-tracking-2015/

https://www.mccaininstitute.org/exploring-the-impact-of-the-super-bowl-on-sex-tracking-2015/

https://arxiv.org/abs/1602.05048

https://arxiv.org/abs/1509.06659

https://arxiv.org/abs/1509.06659

http://dx.doi.org/10.1080/23322705.2015.1015342

http://www.cnn.com/2012/02/29/travel/hotel-sex-tracking/

http://www.cnn.com/2012/02/29/travel/hotel-sex-tracking/

http://www.hungtonpost.com/rachel-lloyd/an-open-letter-to-jim-buc_b_570666.html

http://www.hungtonpost.com/rachel-lloyd/an-open-letter-to-jim-buc_b_570666.html

http://jessicadickinsongoodman.com/2011/10/30/can-we-use-rss-to-catch-rapists-poster-nished/

http://jessicadickinsongoodman.com/2011/10/30/can-we-use-rss-to-catch-rapists-poster-nished/

http://jessicadickinsongoodman.com/2011/10/30/can-we-use-rss-to-catch-rapists-poster-nished/

http://www.disabled-world.com/artman/publish/height-weight-teens.shtml

http://www.disabled-world.com/artman/publish/height-weight-teens.shtml

• Semi-supervised learning for detecting human trafficking

Abstract
Background
Related work
Data preparation
Feature engineering
Advertisement language pattern
Words and phrases of interest
Countries of interest
Multiple victims advertised
Victim weight
Reference to website or spa massage therapy
Unsupervised filtering
Expert assisted labeling
Semi-supervised learning framework
Technical preliminaries
The proposed method
Experimental study
Approaches
Classification results
Blind evaluation
Hyperparameter sensitivity
Significance of features
Conclusion
Authors’ contributions
References