REVIEW

A Review of Data Mining Applications in Crime

Hossein Hassani1∗ , Xu Huang2 , Emmanuel S. Silva3 , and Mansi Ghodsi1
1
Institute for International Energy Studies, Tehran 1967743711, Iran
2
Faculty of Management, Bournemouth University, Fern Barrow, Poole BH125BB, UK
3
Fashion Business School, London College of Fashion, University of the Arts London, London WC1V7EY, UK

Received 23 September 2014; revised 25 February 2016; accepted 28 February 2016

DOI:10.1002/sam.11312
Published online 21 April 2016 in Wiley Online Library (wileyonlinelibrary.com).

Abstract: Crime continues to remain a severe threat to all communities and nations across the globe alongside the
sophistication in technology and processes that are being exploited to enable highly complex criminal activities. Data mining,
the process of uncovering hidden information from Big Data, is now an important tool for investigating, curbing and preventing
crime and is exploited by both private and government institutions around the world. The primary aim of this paper is to provide
a concise review of the data mining applications in crime. To this end, the paper reviews over 100 applications of data mining
in crime, covering a substantial quantity of research to date, presented in chronological order with an overview table of many
important data mining applications in the crime domain as a reference directory. The data mining techniques themselves are
briefly introduced to the reader and these include entity extraction, clustering, association rule mining, decision trees, support
vector machines, naive Bayes rule, neural networks and social network analysis amongst others. © 2016 Wiley Periodicals, Inc.
Statistical Analysis and Data Mining: The ASA Data Science Journal 9: 139–154, 2016

Keywords: big data, data mining, crime, review

1. INTRODUCTION It is in this backdrop that data mining is described as a

Crime has evolved rapidly over time, with criminals
now exploiting the latest in technology not only to
commit crimes but also to evade being captured. Criminal
activity is no longer limited to the streets and back alleys
in our neighborhoods. The Internet, which connects the
entire world, is also a thriving playground for the more
sophisticated criminals in the modern age. Building upon
acts of terror, such as the 9/11 terrorist attacks and use of
technology to hack into the most secure defense databases,
the need for new and effective methods of crime prevention
is increasingly significant [1]. The evolution of ‘Big Data’,
which demands novel approaches towards the effective and
accurate analysis of the growing volumes of crime data,
has been a major challenge for all law enforcement and
intelligence-gathering organizations [2].
∗ niques are further influenced by the increasing availability
Correspondence to: Hossein Hassani (hassani.stat@gmail.com)
powerful tool with great potential to help criminal investiga-
tors focus on the most important information hidden within
the ‘Big Data’ on crime [3]. Data mining as a tool for crime
analysis is recognized as a comparatively new and highly
sought after area of research [4]. This is not surprising as
data mining itself is a relatively new and rapidly evolving
subject, and those interested in the historical and modern
definitions of data mining are referred to [5] as data mining
is not concerned with estimation and tests or prespecified
models but with discovering models through an algorith-
mic search process exploring linear and nonlinear models,
explicit or not.
Alongside the increasing use of the computerized systems
to track crimes, computer data analysts have begun helping
law enforcement officers and detectives speed up the pro-
cess of solving crimes [6] and predict crimes in advance.
The popularity in the application of many data mining tech-

© 2016 Wiley Periodicals, Inc.

140 Statistical Analysis and Data Mining: The ASA Data Science Journal, Vol. 9 (2016)
of Big Data and its ease of use for people who lack data
analysis skills and statistical knowledge [7]. As identified
by many authors, access to data plays an important role
in the effectiveness of data mining in crime, but problems
arise as access is hindered by privacy concerns [2,8].
This paper is aimed at providing a concise review of the
data mining applications used for identifying and prevent-
ing crime over the years. It is expected that this informative
review paper will be useful for introducing Data Mining
techniques to crime researchers and investigators in addition
to supporting and encouraging future research into devel-
oping data mining for crime analysis. In order to enable
such use, the review has been organized so that interested
parties could easily refer to this article alone to apprise
themselves on research that has already been conducted to
date and the resulting outcomes that have been attained. The
main contribution of this paper is two-fold as it not only
captures a majority of the significant data mining appli-
cations in crime by classifying these based on different
types of techniques but also presents a concise introduction
into each of the relevant data mining techniques that have
been exploited for mining crime. Moreover, the review also
includes, in tabular format, a summary of data mining appli-
cations in crime that can act as a quick reference guide for
researchers.
According to ref. [9] the main crime-related data mining
techniques are clustering, association rule mining, classi-
fication and sequential pattern mining. In line with the
findings in ref. [9], our research uncovered that the fol-
lowing data mining techniques are most frequently adopted
for crime analysis. These include entity extraction, cluster-
ing, association rule mining, decision trees, support vector
machines, naive Bayes rule, neural networks and social net-
work analysis.
The remainder of this paper is organized such that a
review of the applications of data mining for crime analysis
is presented in Section 2 in chronological order along
with more specific explanations on the implementation of
data mining techniques. The paper concludes in Section 3,
whilst introductions to various data mining techniques are
presented in the Appendix.
2. DATA MINING APPLICATIONS IN CRIME
In this section, we present a summary of the Data Mining
applications used to detect and prevent crime. In compar-
ison to traditional data mining techniques, the advanced
techniques focus on both structured and unstructured data
to detect patterns [2,10]. With the emergence of Big Data,
most existing systems use a combination of data mining
techniques (which have been introduced in the appendix)
in order to obtain more precise and accurate extractions.
Statistical Analysis and Data Mining: The ASA Data Science Journal DOI:10.1002/sam
In terms of crime, as a broad range of research disciplines,
crime analysis can encompass a wide range of crime activi-
ties, from simple violation of civic duties to internationally
organized crimes [4]. According to ref. [9], complex con-
spiracies are often difficult to unravel because information
on suspects can be geographically diffused and span long
periods of time; detecting cyber crime can likewise be
difficult because busy network traffic and frequent online
transactions generate large amounts of data, whilst only a
small portion would actually relate to illegal activities.
In order to provide a clear view of data mining applica-
tions in crime, Table 1 is provided as a reference directory.
This table summarizes information based on the data mining
technique used and provides information relating to soft-
ware, regions and purpose of the underlying applications.
A recent review of data mining techniques used for finan-
cial accounting fraud detection up until 2011 can be found
in ref. [11] and are therefore not reproduced here.
2.1. Entity Extraction
Entity extraction can be defined as the process of extract-
ing metadata from unstructured text documents.1 Chau et
al. [8] proposed a neural network-based entity extractor,
which applied named-entity extraction techniques to iden-
tify useful entities from police narrative reports in 2002.
They highlighted the importance of valuable information
stored as text objects in criminal-justice data (i.e., the
free-text police narrative reports), which are considered
unstructured data. Unlike the information from structured
data, these unstructured data cannot be easily accessed and
used by investigators or detectives. In ref. [8], four major
named-entity extraction approaches are briefly summarized
and listed accordingly: lexical lookup [98], rule-based [99],
statistic-based [100] and machine learning [98,101,102].
Similar to most existing information extraction systems,
the entity extractor proposed by Chau et al. [8] consists of
more than one of these listed approaches. More specifically,
it combines lexical lookup, machine learning and minimal
hand-crafted rules. By applying the neural network-based
entity extractor in 36 reports randomly selected from the
Phoenix Police Department database for narcotic-related
crimes, the precision rates of extracting entities of persons
and narcotic drug are 74.1% and 85.4%, with recall rates
of 73.4% and 77.9%, respectively [8].
In order to detect what crimes may or may not have been
committed by the same group of individuals, in ref. [12],
a distance measure is proposed with a four-step paradigm
and adaptation of the probability density function to extract
entities from a collection of documents. It is then used to
1
http://www.dataversity.net/entity-extraction-and-the-
semantic-web/
 Table 1. Overview of Data Mining applications in crime.
Data Mining
Techniques References Key Techniques or Softwares
Entity Extraction [8,10,12–20] Named Entity Extraction (lexical lookup,
rule-based, machine learning and
hand-crafted rules), SPSS LexiQuest,
Natural Language Processing, Investigative
Interviewing Technique, Principles of the
Cognitive Interview, General Architecture
for Text Engineering System, Semantic
Inferentialism, Part-of-Speech Tagging,
LBJTatter Algorithm, Conditional Random
Field
Cluster Analysis [6,21–34] GIS (Geographic Information System),
Self-Organizing Map, Hierarchical
Clustering Technique, Partitioning
Clustering Technique, Concept Space
Technique, Co-occurrence Analysis
Algorithm, Link Analysis Technique
(Shortest Path Algorithms,
Priority-First-Search and Two Tree
Priority-First-Search), K-Means Clustering,
Heuristic Approach
Association Rule [40–45] Apriori Algorithm, Distributed Association
Rule Mining (Survey in [46]),Transformed
Categorical Similarities, Dynamically
Adjusted Weights, Outlier Score Function,
Distributed High Order Text Mining,
Temporal Association Rule, Incremental
Mining, Frequent Pattern Growth
Classification [10,32,47–77] Iterative Dichotomiser 3 Algorithm, C4.5
Techniques Algorithm, STAGE Algorithm, CART,
Hunt’s Algorithm, Deceptive Theory
Social Network [78–97] K-core, Core/periphery Ratio; Measure of
Analysis Centrality, Closeness and Betweenness;
Center Weights Algorithm; Borgatti’s Key
Player Approach;
Statistical Analysis and Data Mining: The ASA Data Science Journal DOI:10.1002/sam
Overview Summary
Specific Tested Regions
Phoenix [8], Taiwan [17],
Arabin [18], Malaysian
[19], (New Zealand,
Australia and India) [20]
United Kingdom [35],
Tucson [26,36], Phoenix
[37], Netherlands [27],
Northamptonshire [28],
Tucson [38], India [34,39],
Italy [29], Nigeria [30]
Richmond(Virginia) [40,41],
Hong Kong [43], USA
[44],
Greek [48], USA
[50,51,54,56,74], Asian
[62], Dutch [63], Brazil
[66], Malaysia [76], San
Francisco [77]
Canada [81], United States
[82], Richmond (Virginia,
US) [96]
Purpose and Function
Extract valuable information especially from
unstructured text data (i.e., Person(also Name
by Different Languages), Address, Location,
Time, Vehicle, Nationality, Phone, Gender
and Race, Crime Type, Personal Property,
Organization, Narcotic Drug, Suspect
Descriptions).
Detect crime hot spots; automatically identify
associations from existing crime data and
weight relationships to detect the strongest
association among all possible pairs of crime
related entities; overcome the challenges like
‘information overload, high search complexity
and heavy reliance on domain knowledge’
[38].
Link crime incidents, narrow down possible
suspects, provide informative association
among criminal entities or items, discover
crime patterns.
Efficient detection of specific criminal activities
among large-sized data sets; Categorize crime
data; Predict crime hot spots.
Hassani et al.: Review of Data Mining Applications in Crime
Provide analyses of functions, structures and the
interaction measurements, which, in the crime
domain, detect and illustrate relationships and
connections of criminal entities(including
individuals, groups, events, organizations,
etc.); identify key members and interaction
patterns between sub-groups in criminal
networks; depict the suspects and their
connections to other individuals or
artifacts(like phone numbers, bank accounts).
141
142 Statistical Analysis and Data Mining: The ASA Data Science Journal, Vol. 9 (2016)
transform a high-dimensional vector table into input for a
police-operable tool. By applying this proposed distance
measure, the authors employed the SPSS LexiQuest text
mining tool [103] to form a table of all the entities included
in each investigation. Thereafter, the transformation stage
compared the investigations on common entities, the
adaptation of symmetrical distance measure and probability
density function with the normal distribution, and finally,
a two-dimensional representation of the distances between
all possible couples of investigations is obtained in order to
help the crime analysts and investigators attain an overall
clear understanding of all undergoing investigations.
An Online Crime Reporting System was developed
in ref. [104] to extract relevant crime information from
witness narratives and also to generate additional questions
based on the extracted information. The proposed system
combined natural language processing and an investigative
interviewing technique (based on the cognitive interview
principles) with the adoption of the General Architecture for
Text Engineering System as the information extraction tool.
By evaluating the performance on the suspect description
module, an overall recall rate of 70% and 100% precision
was achieved.
Ku et al. [13,14] recalled that the goal of applying
information extraction techniques in crime analysis is to
help investigators extract crime-related information quickly
and effectively. They developed an online reporting system
that is based on an information extraction technique and
combined natural language processing with insights from
the cognitive interview approach to obtain more information
from witnesses and victims. According to ref. [13], this
proposed system combines information extraction and
principles of the cognitive interview with the purpose of
efficiently gathering more valuable information from those
victims and witnesses who are too scared or embarrassed
to report crime incidents. As this system also encourages
the use of natural language, it not only makes the process
of reporting crime easier but also enables the gathering
of more information. A large lexicon that combines a
rule-based system is developed to extract crime-related
entities by triggering this proposed system to ask questions
according to the principles of the cognitive interview. The
performances of the proposed system show significantly
high precision rates (94% for police narratives and 96%
for witness narratives) and recall rates (85% for police
narratives and 90% for witness narratives).
Pinheiro and Furtado [15] proposed the semantic
inferentialism-based natural language processing model to
extract crime information from unstructured text. In addi-
tion, this framework particularly provided the adaptation of
the collaborative environments on the Web. The evaluation
of this framework was conducted on 100 crime-related texts
on the Web, and a precision rate of 87% was achieved for
Statistical Analysis and Data Mining: The ASA Data Science Journal DOI:10.1002/sam
extracting the crime scene alongside a 72% precision rate
for extracting the type of crime.
In ref. [16], an expanded entity phrase is defined as
the key component for extracting an entity. Effective per-
formance is conducted by combining part of the speech-
based template matching and ontology-driven national
language processing. The preliminary results by imple-
mentation of this proposed approach on free text law
enforcement data outperformed the named-entity extraction
technique and reported over 80% precision and recall rates
in general.
Moreover, Yang et al. [17] used the entity extraction
technique combined with part-of-speech (POS) tagging for
criminal information analysis and relationship visualization.
They reported their method as an efficient and effective
term-relationship mining technique, which showed great
performance even with the most complex case of Chinese
POS tagging during its implementation on criminal infor-
mation data from Taiwan.
In ref. [18], a rule-based Arabic named entity recognition
(NER) system is presented to identify and classify named
entities in Arabian crime text. It comprises of three modules
in the pre-processing stage: sentence splitting, tokenization
and part-of-speech tagging; grammatical rules, patterns and
gazetteer are also considered in the process of the named
entities identification stage. The proposed system achieved
an overall 91% precision rate and 89% recall rate when
tested on the corpus of Arabic crime documents from
newspapers. Recently, Alkaff and Mohd [19] used named
entity recognition with gazetteers and rule-based extraction
for extracting information on nationalities from crime news
in Malaysia.
In ref. [20], NER (LBJTatter algorithm is chosen due to
the best performance in comparison to several algorithms)
was used with conditional random field (a machine-learning
approach to classify a crime location sentence in an
article) to extract crime information found in online news
articles. Both comparisons of two newspapers in New
Zealand and comparisons of crime location extraction
across countries (Australia and India) were conducted with
an overall 80–90% accuracy for tests on New Zealand
newspapers and generally about 75% accuracy for cross-
country scenarios.
2.2. Cluster Analysis
Cluster analysis is a technique used to group observations
where observations within each group are similar, and
the groups are different from each other. Cluster analysis
was combined with the geographic information system
(GIS) for detecting crime hotspots in [25], where the
authors evaluated the performances of different clustering
techniques and stated the concept of cluster significance
 Hassani et al.: Review of Data Mining Applications in Crime
as the standard to define a spot as ‘hot’ or not in crime
analysis.
Adderly and Musgrove [35] combined clustering tech-
niques and self-organizing maps to model the behavior of
sex-offenders in UK. Their results indicated that crimes
within a single cluster have strong similarities and may con-
tain crimes that have been committed by the same offender.
The Coplink system is recognized as one of the
successful implementations of the clustering technique for
crime data mining [2,26,36,37]. According to Hauck et al.
[26], Coplink applies the concept space as its underlying
structure to identify relationships between suspects, victims
and other relevant data in order to accelerate criminal
investigations and enhance law enforcement efforts. To
elaborate, the concept space algorithm is stated as a
statistics-based algorithmic technique that can automatically
compute the strength of relationships between each possible
pair of concept descriptors identified in a document
collection [26]. As most criminal-justice data are properly
structured in the existing reports, the sources that a concept
space can be derived from are located. All the terms in
this located concept space are then filtered and indexed
by an appropriate co-occurrence analysis algorithm. In
ref. [26], terms are categorized into five main types:
Person, Organization, Location, Crime and Vehicle, and
according to feedback from crime analysts and detectives,
Coplink can provide valuable leads by discovering links
between different types of terms and can provide the crucial
advantage of operating efficiency in the investigating
process.
In ref. [105], a link analysis technique is proposed to
identify the strongest association paths between entities in
a criminal network. In comparison to the existing link anal-
ysis tools used in crime analysis (for example, Anacapa
charting system [106], Netmap [107], Analyst’s Notebook
[108], Watson [109] and Coplink Detect [26]), the authors
employed shortest-path algorithms and logarithmic trans-
formation of the link weights to identify the shortest path;
therefore, the strongest associations can be detected among
the crime-related entities.
De Bruin et al. [27] describes a tool that extracts
important factors that play a role in the analysis of criminal
careers from databases and creates digital profiles for all
offenders. Their method compares all individuals on these
profiles using a new distance measure and clusters them
accordingly, which, in turn, yields a visual clustering of
these criminal careers and enables the identification of
classes of criminals [27]. The clustering technique was
employed again in [12] in an investigation process to
identify the offenses that were committed by the same group
of criminals.
K-means clustering was used in [28] to assess the per-
formance of crime scene investigators. By modeling their
Statistical Analysis and Data Mining: The ASA Data Science Journal DOI:10.1002/sam
143
performance, a clear and objective rationale is provided for
further development as a viable path for improving crime
investigation. Schroeder et al. [38] proposed the Crime-
less Explorer system, which combined several techniques,
including co-occurrence analysis, shortest-path algorithm
and the heuristic approach, to provide a better link anal-
ysis to assist crime investigations. In terms of enhancing
investigation effectiveness for Indian police, Gupta et al.
[39] proposed a crime analysis tool with an interactive
query-based interface to help the police in crime investi-
gations. Furthermore, the clustering technique is adopted to
find crime hotspots based on the National Crime Record
Bureau database in India.
More recently, in ref. [29], cluster analysis was used to
identify the effect of various economic factors on crime
in Italy, while in ref. [30], a GIS is complemented with
multivariate cluster analysis for assessing property crime
in Nigeria. Self-organizing map and multilayer perceptron
neural networks are employed with the clustering technique
in ref. [31] for crime analysis and matching. Sukanya
et al. [32] applied the clustering technique with GIS to
identify criminals and crime hotspots by analyzing the
spatial pattern. K-means clustering was used in combination
with the rapid miner tool in ref. [33] for crime analysis of
offences recorded in England and Wales. A spatial-temporal
analysis combining GIS and the clustering technique was
adopted in ref. [34] for day-to-day crime forecasting in
India.
2.3. Association Rule Mining
Association rule mining is a method that exploits the
relationships among observations for uncovering crucial
information hidden within Big Data. In ref. [21], the basic
and advanced concepts and algorithms of association analy-
sis are provided with detailed calculation and examples. In
ref. [110], an association rule mining algorithm is proposed
by using the sales data of customer transactions obtained
from a large retailing company. Association rule mining
is designed for applications where observations consist of
transactions, and a subset of the available items appears in
each transaction. Later, in ref. [111], two new algorithms
are presented and compared with the known algorithms, and
significant improvements are recorded by both new algo-
rithms and their combined AprioriHybrid algorithm.
In ref. [40], association methods are employed for
applications in law enforcement. The proposed approach
automatically looks for similarities by using a new
total similarity measure with information theoretic-based
weights among attributes of robbery records data from the
Richmond Police Department in order to associate incidents
possibly committed by the same or group of criminals.
Thereafter, Lin and Brown presented a new association
144 Statistical Analysis and Data Mining: The ASA Data Science Journal, Vol. 9 (2016)
method that combined outlier score function and tested
it on the same robbery data from the Richmond Police
Department in ref. [41]. The new outlier-based method
outperforms the similarity-based method, with promising
results in providing more helpful information for police
officers.
A distributed higher-order association rule mining algo-
rithm is proposed by Li et al. [112] with example law
enforcement data, which overcomes the requirements of
knowledge of a global schema and the distribution of data
being horizontal or vertical. In ref. [42], association rule
mining is used for detecting suspicious e-mails by identify-
ing unusual and deceptive communication in e-mails with
the implementation of an Apriori Algorithm. This proposed
technique was designed to assist the investigators in effi-
ciently obtaining information and taking effective actions
to reduce or prevent criminal activities. Ng et al. [43]
introduced an incremental algorithm for maintaining tem-
poral association rules with numerical attributes by using
the negative border method. Their new algorithm has been
implemented to support the discoveries of crime patterns in
a district of Hong Kong, and the results showed significant
improvements.
Fuzzy association rule mining was introduced in ref.
[44] as a novel means for knowledge discovery in the
crime domain, supported by experimental results on open-
source communities and crime datasets [113] in USA. The
most significant result is that crime patterns are discovered
and also consistent across all regions, subsets of regions
and all states, which proves the capability and potential
benefit of implementing this proposed technique for crime
analysis and police force investigations. A review of more
applications of association rule mining for crime detection
up until 2014 can be found in ref. [45] and are therefore
not reproduced here.
2.4. Classification Techniques
As one of the most fundamental and significant data
mining techniques, Classification techniques are used
for classifying observations based on some significant
rules/attributes that are discovered from the database. In
terms of applying classification techniques in crime data
mining, many implementations combined more than one
specific type of classification technique. Therefore, the
review that follows is classified by each technique in
chronological order, and depending on circumstances, those
complex combination cases are not reproduced.
2.4.1. Decision Trees
The classification technique, also known as decision
trees, is used in ref. [47] for detecting suspicious e-mails
Statistical Analysis and Data Mining: The ASA Data Science Journal DOI:10.1002/sam
and reported over 95% accuracy in correctly classifying
e-mails in a large-sized dataset. The model of deceptive
theory is applied to the dataset of e-mails, and the
decision tree is generated via the ID3 algorithm. Kirkos
et al. [48] explored and compared the performances of
different classification techniques for auditors in detecting
firms that issue fraudulent financial statements. Three
models (decision trees, neural networks and Bayesian belief
networks) are evaluated by dealing with the identification
of factors associated with fraudulent financial statements on
datasets from 76 Greek manufacturing firms. In ref. [49],
decision trees, neural networks, support vector machines
and stochastic boosting are utilized for hotspot detection in
an urban development project dataset, which contains 1.4
million cases, 14 predictors and a binary response variable.
Decision trees are adopted for predicting crime reporting in
ref. [50] by identifying the variables that influence whether
a crime is reported from survey data obtained via the Bureau
of Justice Statistics of USA, while in ref. [51], classification
techniques like decision trees, neural network, naive Bayes
and support vector machines are applied and compared for
predicting crime hotspots and crime forecasting. In ref.
[52], a reduction strategy of attribute is combined with the
decision trees algorithm for analyzing criminal behavior.
Naive Bayesian, rule-based classification and C4.5 decision
trees are exploited in ref. [53] for detecting auto insurance
fraud. The decision trees technique is particularly compared
with other data mining techniques for automotive insurance
fraud detection in ref. [54]. In ref. [55], a decision tree-
based classification model is used for discovering crime
patterns and predicting future trends. Recently, in ref.
[56], classification algorithms are evaluated for crime
prediction where they find decision trees outperforming
a naive Bayesian algorithm in accurately predicting the
crime category for different states in USA. An improved
decision tree algorithm based on the Maclaurin-Priority
Value First method is used in ref. [57] for computer crime
forensics, and this improved algorithm outperformed the
commonly used ID3 algorithm in terms of both efficiency
and accuracy. More recently, a variety of classification
techniques were evaluated for detecting insurance fraud
in ref. [58] where an algorithm for determining the
relationship between classification models and different
types of insurance fraud data was proposed.
2.4.2. Neural Networks (NN)
Artificial neural networks (ANNs), decision trees and
logistic regression are used in ref. [59] for uncovering
lies from 371 statements of different types of crimes. In
ref. [60], logistic regression and ANNs are applied to
identify smuggling vessels, and the findings indicate that
the ANN reports a higher accuracy than logistic regression.
 Hassani et al.: Review of Data Mining Applications in Crime
Multilayer perceptron (MLP), Decision trees, support vector
machines, genetic programming (GP) and probabilistic
neural networks were compared in ref. [61] for detecting
phishing e-mails from a dataset of 2500 e-mails, whilst in
ref. [62], decision trees, ANNs and support vector machines
were compared for discriminating between those charged
and not charged for initial juvenile offending. Logistic
regression, boosting, SVM, neural networks, K-nearest
neighbor and a variety of other data mining techniques are
evaluated for predicting recidivism in ref. [63].
2.4.3. Support Vector Machines (SVM)
The SVM classification has been used to identify the
sources of e-mail spamming based on the sender’s lin-
guistic patterns and structural features [10,64]. An SVM
model is successfully used in [64] to aid author identifi-
cation forensics via mining e-mail content. SVM is used
for crime hotspot prediction in ref. [65], where they find it
outperforming neural networks and spatial auto-regression-
based approaches. In ref. [66], SVM is exploited for crime
scene classification using a database comprising of 400
crime scenes. The authors find SVM-based approaches per-
forming better than Multilayer perceptron neural networks.
Liu et al. [67] used SVM to help with identifying digital
evidence relating to computer crimes via outlier detection.
In ref. [68], SVM was used to help crime investigators pro-
duce a set of possible interpretations for domain-relevant
concepts through kernel-based relation extraction. Wang et
al. [69] used the SVM technique for predicting criminal
recidivism and compared their results with those obtained
from logistic regression and neural networks. Salem et al.
[70] evaluated the performance of SVM for detecting iden-
tity theft using the Schonlau dataset, and whilst they found
one-class SVM models to be practical in accurately locat-
ing identity theft in the general case, the results were not
satisfactory where specific user profiles fit the average user
profile. Advanced fee fraud activities are detected using
both SVM and random forests (RF) in ref. [71], where
they find SVM outperforming RF2 . Moreover, an SVM
model is used alongside logistic regression and RF for
detecting credit card fraud in [72], where the performances
are evaluated on real world transaction data from inter-
national financial institutions. The SVM technique, com-
bined with the AdaBoost algorithm, is used in ref. [73] for
cyber-crime detection and prevention based on a Facebook
dataset. Alwee et al. [74] exploited a hybrid support vector
regression (SVR) model in combination with ARIMA and
2 Their proposed approach successfully identified and ana-
it is noteworthy that RF have been successfully employed in
ref. [114] in describing the precursors of homicide by accounting
for the relative costs of forecasting errors. The RF algorithms are
also explained in detail in Appendix A of Berk et al. [114].
Statistical Analysis and Data Mining: The ASA Data Science Journal DOI:10.1002/sam
145
particle swarm optimization for forecasting property crime
rates in USA and found it outperforming forecasts from the
individual models, while later in [75], SVR was employed
for forecasting property crime rates along with grey ratio-
nal analysis. The behavior of criminals are analyzed using
SVM in ref. [76] for Malaysia, where the ratio of police to
population is extremely low at 3.6 to 1000. More recently, it
has been shown in ref. [77] that an SVM model can be used
to provide live predictions of crime in urban areas based
on Twitter data related to an urban subarea from within the
city of San Francisco.
2.5. Social Network Analysis (SNA)
Social network analysis (SNA) is a method built on
the analysis of social structure of observations for iden-
tifying significant information. Sparrow [78] summarized
existing concepts of network analysis in the applications of
the crime analysis domain with detailed explanations and
comparisons. Wang and Chiu [79] applied SNA and iden-
tified two transactional network structure measurements,
k-core and core/periphery ratio, to detect the online auction
inflated-reputation traders from regular accounts. Signifi-
cant results proved that SNA can act as an effective indictor
to distinguish criminal accounts and possibly prevent and
reduce problematic transactions and online auction frauds.
Qin et al. [80] proposed an analysis of the structure of the
Global Salafi Jihad network with SNA and Web structural
mining. Results showed that the proposed technique can be
an effective tool to identify key members in a terrorist net-
work and, therefore, help authorities develop efficient and
effective disruptive strategies and measures.
In ref. [81], the SNA technique was used to investi-
gate organized criminal groups by applying it to an outlaw
motorcycle gang operating in Canada. Ressler [82] intro-
duced SNA as a new type of intelligence for homeland
security in USA, which can provide important informa-
tion on the unique characteristics of terrorist organizations,
understand terrorist networks and form the basis for a more
effective countermeasure to net war. In ref. [83], SNA was
employed on the Enron corporation’s e-mail archive, and it
proved to be able to analyze and catalog patterns of commu-
nications between entities in an e-mail collection to extract
social standing. The authors stated that this technique makes
it possible to view a snapshot of a corporate community and
effectively determine the real relationships and connections
between individuals.
Chau and Xu [84] applied the web mining and network
analysis technique to study hate groups in online blogs.
lyzed a selected set of hate groups (which contained 820
bloggers) on Xanga. Nair and Sarasamma [85] performed
SNA alongside fuzzy theory with the aim of modeling
146 Statistical Analysis and Data Mining: The ASA Data Science Journal, Vol. 9 (2016)
multi-modal social networks. A new fuzzy binary operation
was proposed to satisfy the requirements of a fuzzy con-
solidation operator. Liu et al. [86] proposed an enhanced
shortest-path algorithm, combing SNA to mine the core
member of a terrorist group. Wang and Chiu [77] used two
SNA indicators, k-core and center weights algorithms, to
form a recommendation system that can suggest the risks
of collusion associated with an account for online auction
site users. The results are promising, with 76% detection
accuracy on real world ‘blacklist’ data and sound evidence
that the approach can provide effective warnings several
months ahead of the official release of blacklists.
In ref. [88] the authors discussed the solution for the
case of privacy-preserving SNA in collaboration with multi-
ple investigators. The proposed algorithm, compute metrics,
can operate the computation and analysis without knowing
the personally identifiable data with high privacy guaran-
tees. Chau and Xu [89] again addressed the importance of
studying the emergence of cyber communities in blogs with
a combination of web mining and SNA techniques. In ref.
[90], the authors designed a simulation e-mail system based
on personality trait dimensions to model the traffic behav-
ior of e-mail account users and employed this with SNA
to identify key members of a criminal group. Chen [91]
shows the applicability of SNA for homeland security data
mining with case studies based on gang/narcotic networks,
US extremist networks, Al-Qaeda member networks and
international Jihadist websites and forum networks.
Fard and Ester [92] proposed a framework for automated
network data analysis and deduction from multiple social
networks by converting a transaction dataset and applying
association mining and statistical methods. The proposed
method varies from previous work and combined the game
theory concept in a multi-agent model with the purpose
of building P2P applications for the police force to detect
relationships between criminals and narrow down possible
suspects. In ref. [93], the authors proposed an SNA-
based model for targeting criminal networks. It is built
on Borgatti’s key player approach with modifications on
incorporating the relative strength of actors as well as the
strength of the relationships binding network actors. Chiu
et al. [94] employed SNA for identifying internet auction
fraud. They performed experiments on data from the Yahoo
Auctions website with comparisons of different types of
internet auction accounts and achieved promising results
pertaining to prediction accuracy.
In ref. [95], Carrington classified the applications of
SNA in the crime domain into three areas (the influence
of the personal network on ego’s delinquency or crime,
the influence of neighborhood networks on crime in the
neighborhood and the organization of criminal groups
and activities) and provided detailed explanations together
with significant theories and literature. The SNA technique
Statistical Analysis and Data Mining: The ASA Data Science Journal DOI:10.1002/sam
was used as a tool for helping crime analysts and
detectives in police forces develop interdiction strategies
in ref. [96], with an example case study focusing on the
Richmond City Police Department. Detailed examples of
implementations are provided to show how the proposed
approach can assist police in understanding complex
behavioral motivations of offenders, strategically hot-
spotting people of interest and developing stronger inter-
jurisdictional working relationships [96]. More recently,
in ref. [97], SNA was used to detect and analyze hidden
activities in social networks. Both regression-based and
Min Cut-based algorithms are adopted and compared with
datasets from several sources.
3. CONCLUSION
This review paper begins with a brief introduction into
the evolution of crime and the importance of data min-
ing for crime detection, analysis and prevention. Given the
vast amount of funding that is allocated for defense-related
expenditure around the globe, it is clear that considerable
savings could be attained through the correct application
of data mining techniques, which are primarily aimed at
uncovering hidden relationships in Big Data. Following
thorough research, we are able to present a list of data min-
ing techniques as the most frequently adopted at present for
crime analysis. These include entity extraction, clustering,
association rule mining and classification techniques like
decision trees, support vector machines, naive Bayes rule,
neural networks and social network analysis.
As opposed to providing a review of applications alone,
this paper goes a step further and provides a concise intro-
duction to these data mining techniques (see Appendix) as
well as a brief summary of the specific functions of each
technique in crime analysis and thereby enables readers
from different backgrounds to obtain a much richer under-
standing of the underlying process whilst guiding them to
relevant articles for a more detailed description. Table 1
particularly functions as a useful resource or ‘quick guide’,
which summarizes the data mining applications in crime,
providing useful information not only on the software and
purpose but also the regions that have been tested.
This review itself is unique as it captures over 100
applications and is the most up-to-date and thorough review
of data mining applications in crime to date. We find
strong evidence based on the number of applications to
conclude that classification techniques are the most popular
form of data mining in crime. This is interesting as there
exists a major difference between the findings from data
mining in official statistics [5] and our findings here.
We notice that SVM, neural networks and association
rule mining are seldom used to mine official statistics,
 Hassani et al.: Review of Data Mining Applications in Crime
whereas in crime data mining, these methods are extremely
popular and well exploited. This further highlights the
institutional and cultural differences that exist between
national statistical institutes and organizations engaged in
analyzing crime.
The increasing applications of data mining techniques
alongside the emergence of Big Data also points toward
the need for more training and investment in educating
and empowering youth with knowledge on the advantages,
developments and practical use of data mining techniques.
In countries such as the UK, it is encouraging to see
universities now offering not only units but also courses
on data mining, which will most certainly have a positive
impact on future generations as well as lead towards
an exponential increase in data mining applications. We
believe that this review paper can help researchers around
the globe easily identify existing research so as to develop
more complex and improved techniques for mining crime
data in the future.
APPENDIX
A. DATA MINING TECHNIQUES IN CRIME
This appendix aims to give the reader an introduction to the data mining
techniques that have been used for crime analysis.
A.1 Entity Extraction
Entity extraction is a process that depends on the availability of
extensive amounts of clean input data for the identification of particular
patterns such as text, images or audio materials that can then provide
basic information for crime analysis [2,10]. The Message Understanding
Conference (MUC) series, which coordinates multiple research groups and
government agencies [115], has been the major forum for researchers in
this area, where they meet and compare the performance of their entity
extraction approaches [116] and seek to improve the technologies [115].
As concluded in ref. [8], the main approaches for entity extraction are
lexical-lookup, rule-based, statistic-based and machine learning. Below,
we provide some basic introductions to the major approaches, and in doing
so, we mainly follow ref. [8].
Lexical lookup Most entity extraction systems maintain hand-crafted
lexicons that contain lists of popular names for entities of interest.
These systems work by looking up phrases in text that match the
items specified in their lexicons. See, for example, [98].
Rule-based The rule-based system relies on hand-crafted rules to identify
named entities. The rules may be structural, contextual or lexical
[99].
Statistic-based Statistical-based systems often use statistical models to
identify occurrences of certain cues of particular patterns for
entities in texts. A training dataset is needed for these systems
to obtain the statistics. The statistical language model in ref. [100]
is an example of statistical-based entity extraction systems.
Machine learning Instead of human-created rules, the machine learning
system relies on what is termed as machine learning algorithms to
extract knowledge and identify patterns from texts. Few examples
Statistical Analysis and Data Mining: The ASA Data Science Journal DOI:10.1002/sam
147
are neural networks, decision trees [101], hidden Markov models
[102] and entropy maximization [98].
A.2 Cluster Analysis
According to ref. [21], cluster analysis groups data objects based only
on information found in the data that describes the objects and their
relationships. The key objective of this process is to find groups of objects
such that the objects in a group will be similar (or related) to one another
and different from (or unrelated to) the objects in other groups [21]. The
fundamental concept of clustering is the distance measurement among
objects. Euclidean distance, Manhattan distance and Minkowski distance
are the most widely adopted distance measures for cluster analysis. In
order to identify different types of structures or clusters in the data,
different algorithms have been developed to date, which can be mainly
classified into hierarchical methods and partitioning methods based on
how the subsets are organized [24]. Other methods, except the two
major approaches, include the density-based method, grid-based method,
model-based method and constraint-based method. For the sake of briefly
introducing the process of cluster analysis without covering all available
algorithms to date, we introduce one specific algorithm of cluster analysis
below by following ref. [22]. Note that alternatives exist for each step due
to the significant developments in cluster analysis. More details can be
found in refs. [21–24].
Given a set of points S = {s1 , ..., sn } in Rl , which we wish to cluster
into k subsets, the first stage will be the ordination of the data (includes
steps 1–4) followed by the second stage, which focuses on clustering
(includes steps 5–6):
1. From the affinity matrix A ∈ Rn×n defined by Aij = exp(−
 si − sj 2 /2σ 2 ) if i = j , and Aij = 0. Note that  si − sj 
refers to the normalized distance between si and sj .
2. Define D as the diagonal matrix whose (i, i)-element is the sum
of A’s i-th row, and construct the matrix L = D−1/2 AD−1/2 .
3. Find x1 , x2 , ..., xk , the k largest eigenvectors of L (chosen to be
orthogonal to each other in the case of repeated eigenvalues)
and form the matrix X = [x1 x2 ...xk ] ∈ Rn×k by stacking the
eigenvectors in columns.
4. Form the matrix Y from X by re-normalizing each of X’s rows to
have unit length (i.e., Yij = Xij /(j Xij2 )1/2 ).
5. Treating each row of Y as a point in Rk , cluster them into k clusters
via K-means or any other algorithm (that attempts to minimize
distortion).
6. Finally, assign the original point si to cluster j if and only if row
i of the matrix Y was assigned to cluster j .
A.3 Association Rule Mining
Association rule mining was initially proposed by Agrawal et al. [110]
as a method of discovering interesting co-occurrences in supermarket
data. The association rule is defined as a technique for investigating the
possibility of the simultaneous occurrence of data [117]. Yun et al. [117]
suggested an association rule discovery technique that is able to identify
significant, rare data associated with specific data in a way that the rare
data occurs simultaneously with the specific data more frequently than the
average co-occurrence frequency in the database. We follow ref. [111] and
148 Statistical Analysis and Data Mining: The ASA Data Science Journal, Vol. 9 (2016)

briefly introduce the association rule mining technique below. For a more
detailed explanation, see ref. [110].
Let I = {i1 , i2 , ..., im } be a set of literals, called items. Let D be a set
of transactions, where each transaction T is a set of items such that T ⊆ I .
Associated with each transaction is a unique identifier, called T I D. We
say that a transaction T contains X, a set of some items in I , if X ⊆ T .
An association rule is an implication of the form X ⇒ Y , where X ⊂ I ,
Y ⊂ I and X ∩ Y = ∅.
More specifically, in ref. [21], the strength of an association rule can
be measured in terms of its support and confidence, in which support
determines how often a rule is applicable to a given dataset, while
confidence determines how frequently items in Y appear in transactions
that contain X. The formal definitions of these measures are as follows,
in which N refers to the number of transactions:
σ (X ∪ Y )
Support,s(X ⇒ Y ) = . (A1)
N
σ (X ∪ Y )
Confidence,c(X ⇒ Y ) = . (A2)
σ (X)
Therefore, it can be concluded that support indicates that the rule
X ⇒ Y holds in the transaction set D with confidence c if c% of
transactions in D contain X ∪ Y ; and confidence indicates that the rule
X ⇒ Y holds in the transaction set D with confidence c if c% of
transactions in D that contain X also contain X ∪ Y . Given a set of
transactions D, the problem of association rule mining is to generate
all association rules in the database that have support and confidence
greater than the user-specified minimum support (called minsup) and
minimum confidence (called minconf ), respectively [111]. There are many
algorithms for association rule mining, among which, the most well-known
and widely-used Apriori algorithm was proposed in refs. [110,111] as a
seminal algorithm for identifying support using candidate generation. Wu
et al. [118] stated that the Apriori algorithm is one of the most popular data
mining approaches to find support from a transaction dataset and derive
association rules. Other algorithms, like direct hashing and pruning (DHP),
FP-growth and H-mine, can be found with more details in [119–122].
A.4 Classification Techniques
records are summarized in a table named a confusion matrix. Then, the
performance metric, which summarizes the information in the confusion
matrix into a single number, provides an easier way for comparing the
performance of different classification models [21].
Decision Trees
Decision trees ( [124–126]) are applied to accomplish the classification
task by giving a series of carefully crafted questions about the attributes
of the test record [21]. When an answer is achieved, it will be followed
by a question until the category of this attribute is concluded. All the
series of questions and possible answers are carefully predefined as well
as the process repeated to all subsets of the tree. The classification task
is established at the same time a decision tree is constructed. Note that
the training records are split based on an attribute test that optimizes
some certain criterions; additionally, it is possible that many different
decision trees can be conducted from the same given set of cases. The final
choice depends on the research and the individual circumstances. Many
algorithms have been developed for getting the most reasonably optimal
decision tree with good accuracy in a timely manner. For example, CART
[124], C4.5 [125], ID3 [126], Hunt’s Algorithm [127], SLIQ [128] and
SPRINT [129].
Among all the effective algorithms, Hunt’s algorithm is the basis of
many existing decision tree-induction algorithms, by which the decision
tree is constructed via recursively partitioning until each path ends in a
pure subset [21]. The process of Hunt’s algorithm is listed below, mainly
following [21]. In addition, one example problem of predicting whether
a loan applicant will succeed in repaying her loan obligations or become
delinquent is adopted for applying Hunt’s algorithm and inducing decision
trees in Figure A2. Let Dt be the set of training records that are associated
with node t, and y = y1 , y2 , ..., yc be the class labels.
Step One If all the records in Dt belong to the same class yt , then t is
a leaf node (which refers to a node that has exactly one incoming
edge and no outgoing edges), and it is labeled as yt .
Step Two If Dt contains records that belong to more than one class,
an attribute test condition is selected to recursively partition the
records into smaller subsets. A child node is created for each
outcome of the test condition, and the records in Dt are distributed

As one of the most fundamental and significant data mining techniques,

classification is briefly defined in ref. [21] as the task of assigning objects
to one of several predefined categories. The aim of classification rule
mining is to discover a small set of rules in the database to form an accurate
classifier [123]. The formal description of the algorithm underlying the
classification rule is listed below by mainly following ref. [21].
The input data for a classification task can be regarded as a collection of
records that contains the instances that are characterized by a tuple (X, y)
(note that X is the attribute set, y is a special attribute, also known as
category). Therefore, classification can be explained as the task of learning
a target function f that maps each attribute set X to one of the predefined
class labels y (see Figure A1). In order to evaluate the performance of
a classification model, the counts of correct and incorrect predicted test

Fig. A1 Classification as the task of mapping an input attribute Fig. A2 Example of Hunt’s algorithm for inducing decision trees
set X into its class label y [21]. [21].

Statistical Analysis and Data Mining: The ASA Data Science Journal DOI:10.1002/sam
 Hassani et al.: Review of Data Mining Applications in Crime
to the children based on the outcomes. The algorithm is then
recursively applied to each child node.
Support Vector Machines
Support vector machines (SVM) were initially introduced in ref. [130]
as a training algorithm for classification, which divides the objects into
two classes by using an optimal separating hyperplane that minimizes
the classification error (see Figure A3). We briefly present the theory
underlying SVM below by following [130].
Consider the simplest problem of separating a set of training vectors
D belonging to two separate classes:
D = {(x 1 , y 1 ), ..., (x l , y l )}, where xεR n , yε{−1, 1}. (A3)
The hyperplane that geometrically separates the surfaces of two classes is
defined as < w, x > +b = 0, where w is the weight vector and b refers to
the bias. Additionally, the parameters are constrained by mini | < w, x i >
+b| = 1 [130]. A separating hyperplane in such canonical form must
satisfy the following constraint: y i [< w, x i > +b] ≥ 1, where i = 1, ..., l.
The set of vectors is said to be optimally separated by the hyperplane
if it is separated without error, and the distance d between the closest
vectors to the hyperplane is maximal, which can be computed by
| < w, x i > +b|
d(w, b, x) = . (A4)
||w||
The optimal separating hyperplane is attainable via maximizing ρ (the
margin) subject to the constraints in the canonical form of the hyperplane
2 fnb (E) =
ρ(w, b) = min d(w, b, x i ) + min d(w, b, x i ) = , (A5)
x i :y i =−1 x i :y i =1 ||w||
and minimizes
1
φ(w) = ||w||2 . (A6)
2
Naive Bayes Rule
The naive Bayes classifier was proposed in Langley et al. [131] and uses
Bayes Rule to compute the probability of each class given the instance,
Fig. A3 Classification by Support Vector Machines [130].
Statistical Analysis and Data Mining: The ASA Data Science Journal DOI:10.1002/sam
149
assuming the attributes are conditionally independent given the label [132].
In general, it is simple and easy to understand; furthermore, it is also
convenient for implementation. This classifier is recognized as one of
the most efficient and effective inductive learning algorithms for machine
learning and data mining [133]. We follow ref. [133] to present a detailed
algorithm of the naive Bayes rule.
Assume an example set E is represented by a set of attribute values
(x1 , x2 , ..., xn ), where xi is the value of attribute Xi . Let C represent the
classification variable, and let c be the value of C. Note that here, we
assume that there are only two classes: + (positive class) or - (negative
class).
According to Bayes Rule, the probability of an example E being class
c is:
p(E | c)p(c)
p(c | E) = , (A7)
p(E)
where E is classified as the class C = + if and only if
p(C = + | E)
fb (E) = ≥ 1, (A8)
p(C = − | E)
where fb (E) is called a Bayesian classifier.
Assume that all attributes are independent given the value of the class
variable, that is:

n
p(E | c) = p(x1 , x2 , ..., xn | c) = p(xi | c). (A9)
i=1
Then the resulting classifier is:
p(C = +)  p(xi | C = +)
n
. (A10)
p(C = −) p(xi | C = −)
i=1
The function fnb (E) is so called a naive Bayesian classifier (or Naive
Bayes). Therefore, the p(x1 | c), p(x2 | c), ..., p(xn | c) can be estimated
from a training sample, and the posterior probability for each class can
be calculated respectively. The class with the highest posterior probability
will be the prediction class.
Neural Networks
As one of the most important tools for classification, neural networks
has proved its high tolerance to noisy data and the ability to classify
untrained patterns with convincing evidence through recently published
research. According to ref. [134], neural networks are able to estimate
the posterior probabilities, which provides the basis for establishing
classification rule and performing statistical analysis [134–138]. Effective
and successful performances have been achieved by neural networks
in many real-world classification tasks [139]. The algorithm of neural
networks is introduced below by mainly following [135].
Consider a general M-group classification problem in which each object
has an associated attribute vector x of d dimensions. Note that ω denotes
the membership variable that takes a value of ωj if an object belongs to
group j . Neural network for a classification problem can be viewed as a
mapping function
F : Rd → RM , (A11)
where d-dimensional input x is submitted to the network, and an
M-vectored network output y is obtained to make the classification
decision. The network is built so that an overall error measure, such
150 Statistical Analysis and Data Mining: The ASA Data Science Journal, Vol. 9 (2016)
as the mean squared error, is minimized. According to ref. [135], the
mapping function F : x → y that minimizes the expected squared error
E[y − F (x)]2 is the conditional expectation of y given x F (x) = E[y | x].
Hence, the j th element of F (x) is given by
Fj (x) = E[yj | x]
= 1 · P (yj = 1 | x) + 0 · P (yj = 0 | x)
(A12)
= P (yj = 1 | x)
= P (ωj | x).
That is, the least squares estimate for the mapping function in a
classification problem is exactly the posterior probability. The mean
squared error function can be derived as
M 
 Workshops, Sydney, 2008, 8–11.
MSE = [Fj (x) − P (ωj | x)]2 f (x)dx
j =1 Rd
M 
 National Conference on Digital Government Research,
+ P (ωj | x)(1 − P (ωj | x))f (x)dx. (A13)
Rd
j =1
The first term of the right side, termed as the estimation error, is affected
by the effectiveness of neural network mapping, while the second term is
independent of neural networks and reflects the inherent irreducible error
due to randomness of the data [135].
A.5 Social Network Analysis
Social network data is now a significant resource for criminal
investigations. According to Mena [140], social network analysis (SNA)
is a data mining technique that reveals the structure and content
found in a body of information by representing the same as a set
of interconnected, linked objects. The basic structure of a social
network consists of nodes that are connected to other related nodes
by links. The connections between two nodes is called an edge, which
indicates that two persons participated in a certain event [141]. The
measurement techniques used in social network analysis are based on
the principles of graph theory, which consists of various mathematical
formulae and concepts for the study of pattern of lines [141] (for
more details see, [143–145]). Among all the measurement techniques,
degree, density and centrality are the most significant and widely used
techniques.
Degree The ‘degree’ of any node of the network is defined as the
number of other nodes to which it is directly linked [78]. It simply
represents the scores or count of the number of ties with other
actors in the network.
Density Density is defined as the number of edges in a portion of a social
network to the maximum number of edges that theoretically make
up the social network [104]. Assume C is the number of observed
edges and n is the total number of nodes in the social network,
then the formula of density is, Density = n(n−1)/2
C
. [14] C. H. Ku, A. Iriberri, and G. Leroy, Natural language
Centrality According to ref. [146], centrality is the measure of closeness
of a node to the center of high activity in a network, which
indicates the structural importance level of the node. The centrality
can be evaluated by degree, closeness and betweenness: Degree
centrality refers to the direct count of the number of connections a
node has to other nodes; Closeness centrality is considering the
overall closeness of a node to all other nodes in the network
[147]; Betweenness centrality measures the extent of a node’s
Statistical Analysis and Data Mining: The ASA Data Science Journal DOI:10.1002/sam
placement on the shortest path between other pairs of nodes in the
network [145].
REFERENCES
[1] P. Kanellis, E. Kiountouzis, N. Kolokotronis, and D. Mar-
takos, Digital Crime and Forensic Science in Cyberspace,
Hershey, PA, IGI Global, 2006.
[2] G. K. Gupta, Introduction to Data Mining with case studies,
New Delhi, Prentice Hall, 2006.
[3] P. Thongtae and S. Srisuk, An analysis of data mining
applications in crime domain, In IEEE 8th International
Conference on Computer and Information Technology
[4] H. Chen, W. Chung, Y. Qin, M. Chau, J. J. Xu, G. Wang, R.
Zheng, and H. Atabakhsh, Crime data mining: an overview
and case studies, In Proceedings of the 2003 Annual
Boston, MA, 2003, 1–5.
[5] H. Hassani, G. Saporta, and E. S. Silva, Data mining and
official statistics: the past, the present and the future, Big
Data 2 (1) (2014), 34–43.
[6] S. V. Nath, Crime pattern detection using data mining,
In Proceedings of the International Conference on Web
Intelligence and Intelligent Agent Technology Workshops,
Hong Kong, 2006, 41–44.
[7] U. Fayyad and R. Uthurusamy, Evolving data into mining
solutions for insights, Commun ACM 45 (8) (2002),
28–31.
[8] M. Chau, J. J. Xu, and H. Chen, Extracting meaningful
entities from police narrative reports, In Proceedings of the
2002 Annual National Conference on Digital Government
Research, Los Angeles, CA, 2002, 1–5.
[9] J. Hosseinkhani, M. Koochakzaei, S. Keikhaee, and Y. H.
Amin, Detecting suspicion information on the web using
crime Data Mining techniques, Int J Adv Comput Sci
Inform Technol 3 (1) (2014), 32–41.
[10] H. Chen, W. Chung, J. J. Xu, G. Wang, Y. Qin, and M.
Chau, Crime data mining: a general framework and some
examples, Computer 37 (4) (2004), 50–56.
[11] A. Sharma, and P. K. Panigrahi, A review of financial
accounting fraud detection based on data mining tech-
niques, Int J Comput Appl 39 (1) (2012), 37–47.
[12] T. K. Cocx, and W. A. Kosters, A distance measure
for determining similarity between criminal investigations,
Adv Data Mining Appl Med Web Mining Marketing Image
Signal Mining 4065 (2006), 511–525.
[13] C. H. Ku, A. Iriberri, and G. Leroy, Crime information
extraction from police and witness narrative reports, In
Proceedings of the IEEE Conference on Technologies for
Homeland Security, 12–13 May, Waltham, MA, 2008,
193–198
processing and e-government: crime information extraction
from heterogeneous data sources, In Proceedings of the
9th Annual International Digital Government Research
Conference, Montreal, Canada, 2008, 18–21.
[15] V. Pinheiro, V. Furtado, T. Pequeno, and D. Nogueira, Nat-
ural language processing based on semantic inferentialism
for extracting crime information from text, In Proceedings
of the IEEE International Conference on Intelligence and
Security Informatics (ISI), Vancouver, BC, 2010, 23–26.
 Hassani et al.: Review of Data Mining Applications in Crime
[16] J. Johnson, A. Miller, L. Khan, B. Thuraisingham, and
M. Kantarcioglu, Extraction of expanded entity phrases,
In Proceedings of the IEEE International Conference on
Intelligence and Security Informatics (ISI), Beijing, China,
2011, 10–12.
[17] K.-S. Yang, C.-C. Chen, Y.-H. Tseng, and Z.-P. Ho, Name
entity extraction based on POS tagging for criminal infor-
mation analysis and relation visualization. In Proceedings
of the 6th International Conference on New Trends in
Information Science and Service Science and Data Mining
(ISSDM), 23–25 October, Taipei, Japan, 2012, 785–789.
[18] M. Asharef, N. Omar, and M. Albared, Arabic named
entity recognition in crime documents, J Theor Appl Inform
Technol 44 (1) (2012), 1–6.
[19] A. Alkaff, and M. Mohd, Extraction of nationality from
crime news, J Theor Appl Inform Technol 54 (2) (2013),
304–312.
[20] R. Arulanandam, B. T. R. Savarimuthu, and M. A. Purvis,
Extracting crime information from online newspaper arti-
cles, In Proceedings of the Second Australasian Web Con-
ference, 20–23 January, Auckland, New Zealand, 2014,
31–38.
[21] T. Pang-Ning, M. Steinbach, and V. Kumar, Introduction
to Data Mining, (1st ed.), London, Pearson, 2014.
[22] T. G. Dietterich, S. Becker, and Z. Ghahramani, Advances
in neural information processing systems 14, In Proceed-
ings of the Annual Conference on Neural Information Pro-
cessing Systems, MIT Press, 2002.
[23] M. R. Anderberg, Cluster analysis for applications (No.
OAS-TR-73-9). Office of the Assistant for Study Support
Kirtland Afb N Mex, 1973.
[24] R. T. Ng and J. Han, Efficient and effective clustering
methods for spatial data mining, In Proceedings of
The 20th International Conference on Very Large Data
Bases, September 12–15, Chile, Santiago de Chile, 1994,
144–155.
[25] T. Grubesic, and A. Murray, Detecting Hot-spots using
Cluster Analysis and GIS, In Proceedings of the 5th Annual
International Crime Mapping Research Conference, Dallas,
2001.
[26] R. V. Hauck, H. Atabakhsb, P. Ongvasith, H. Gupta, and
H. Chen, Using Coplink to analyze criminal-justice data,
Computer 35 (3) (2002), 30–37.
[27] J. S. De Bruin, T. K. Cocx, W. A. Kosters, J. F. J. Laros,
and J. N. Kok, Data Mining approaches to criminal career
analysis, In Proceedings of the 6th International Conference
on Data Mining, Hong Kong, 2006, 18–22.
[28] R. Adderley, M. Townsley, and J. Bond, Use of data mining
techniques to model crime scene investigator performance,
Knowl Based Syst 20 (2) (2007), 170–176.
[29] R. Lombardo, and M. Falcone, 2011. Crime and Economic
Performance. A cluster analysis of panel data on Italy’s
NUTS 3 regions. Working Paper, University of Calabria,
1–33.
[30] Y. Bello, and S. A. Yelwa, Complementing GIS with
Cluster Analysis in Assessing Property Crime in Katsina
State, Nigeria, Am Int J Contemp Res 2 (7) 2012, 190–198.
[31] M. R. Keyvanpour, M. Javideh, and M. R. Ebrahimi,
Detecting and investigating crime by means of data mining:
a general crime matching framework, Procedia Comput Sci
3 (2011), 872–880.
[32] M. Sukanya, T. Kalaikumaran, and S. Karthik, Criminals
and crime hotspot detection using data mining algorithms:
Statistical Analysis and Data Mining: The ASA Data Science Journal DOI:10.1002/sam
151
clustering and classification, Int J Adv Res Comput Eng
Technol 1 (10) (2012), 225–227.
[33] J. Agarwal, R. Nagpal, and R. Sehgal, Crime analysis using
K-means clustering, Int J Comput Appl 83 (4) (2013), 1–4.
[34] M. Vijayakumar, S. Karthick, and N. Prakash, The day-
to-day crime forecasting analysis of using spatial-temporal
clustering simulation, Int J Sci Eng Res 4 (1) (2013), 1–6.
[35] R. Adderley, and P. B. Musgrove, Data mining case study:
modeling the behavior of offenders who commit serious
sexual assaults, In Proceedings of the 7th International
Conference on Knowledge Discovery and Data Mining,
San Francisco, CA, 2001, 26–29.
[36] H. Chen, J. Schroeder, R. V. Hauck, L. Ridgeway,
H. Atabakhsh, H. Gupta, C. Boarman, K. Rasmussen,
and A. W. Clements, COPLINK connect: information
and knowledge management for law enforcement, Decis
Support Syst 34 (3) (2002), 271–285.
[37] H. Chen, D. Zeng, H. Atabakhsh, W. Wyzga, and J.
Schroeder, COPLINK: managing law enforcement data and
knowledge, Commun ACM 46 (1) (2003), 28–34.
[38] J. Schroeder, J. Xu, H. Chen, and M. Chau, Automated
criminal link analysis based on domain knowledge, J Am
Soc Inform Sci Technol 58 (6) (2007), 842–855.
[39] M. Gupta, B. Chandra, and M. P. Gupta, Crime data mining
for Indian police information system. Computer Society of
India, 2008, 389–397.
[40] D. E. Brown, and S. Hagen, Data association methods with
applications to law enforcement, Deci Support Syst 34 (4)
(2003), 369–378.
[41] S. Lin, and D. E. Brown, An outlier-based data association
method for linking criminal incidents, Decis Support Syst
41 (3) (2006), 604–615.
[42] S. Appavu, M. Pandian, and R. Rajaram, Association
rule mining for suspicious email detection: a data mining
approach, In Proceedings of the IEEE Intelligence and
Security Informatics, Brunswick, NJ, 2007, 23–24.
[43] V. Ng, S. Chan, D. Lau, and C. M. Ying, Incremental
mining for temporal Association Rules for crime pattern
discoveries, In Proceedings of the 18th Conference on
Australasian Database, Ballarat, Victoria, 2007, 29 January-
2 February, 123–132.
[44] A. L. Buczak, and C. M. Gifford, Fuzzy association
rule mining for community crime pattern discovery,
In Proceedings of the ACM SIGKDD Workshop on
Intelligence and Security Informatics, Washington, DC,
2010, 25–28.
[45] D. Usha, and K. Rameshkumar, A complete survey on
application of frequent pattern mining and association rule
mining on crime pattern mining, Int J Adv Comput Sci
Technol 3 (4) (2014), 264–275.
[46] M. J. Zaki, Parallel and distributed association mining: a
survey, IEEE Concurrency 7 (4) (1999), 14–25.
[47] S. Appavu, and R. Rajaram, Suspicious e-mail detection
via decision tree: a data mining approach, J Comput Inform
Technol 15 (2) (2007), 161–169.
[48] E. Kirkos, C. Spathis, and Y. Manolopoulos, Data Mining
techniques for the detection of fraudulent financial state-
ments, Expert Syst Appl 32 (4) (2007), 995–1003.
[49] C. Wang, and P.-S. Liu, Data mining and hotspot detection
in an urban development project. J Data Sci 6 (2008),
389–414.
[50] J. Gutierrez, Using decision trees to predict crime reporting,
In Advanced Principles for Improving Database Design,
Systems Modeling, and Software Development, K. Siau
152 Statistical Analysis and Data Mining: The ASA Data Science Journal, Vol. 9 (2016)
and J. Erickson, eds. Hershey, PA, Information Science
Reference, 2009, 132–145.
[51] C.-H. Yu, M. W. Ward, M. Morabito, and W. Ding, Crime
forecasting using data mining techniques. In Proceedings
of the 11th International Conference on Data Mining
Workshops, 11 December, Vancouver, BC, 2011, 779–786.
[52] W. Hui, W. Jing, and Z. Tao, Analysis of decision tree
classification algorithm based on attribute reduction and
application in criminal behavior. In Proceedings of the
3rd International Conference on Computer Research and
Development, 11–13 March, Shanghai, 2011, 27–30.
[53] R. Bhowmik, Detecting auto insurance fraud by data
mining techniques, J Emerg Trends Comput Inform Sci
2 (4) (2011), 156–162.
[54] A. Gepp, J. H. Wilson, K. Kumar, and S. Bhattacharya,
Comparative analysis of decision trees vis-à-vis other com-
putational data mining techniques in automotive insurance
fraud detection, J Data Sci 10 (2012), 537–561.
[55] A. Nasridinov, S.-Y. Ihm, and Y.-H. Park, A decision
tree-based classification model for crime prediction, Inform
Technol Convergence Lect Notes Electr Eng 253 (2013),
531–538.
[56] R. Iqbal, M. A. A. Murad, A. Mustapha, P. H. S.
Panahy, and N. Khanahmadliravi, An experimental study
of classification algorithms for crime prediction, Ind J Sci
Technol 6 (3) (2013), 4219–4225.
[57] Y. Wang, X. Peng, and J. Bian, Computer crime forensics
based on improved decision tree algorithm, J Netw 9 (4)
(2014), 1005–1011.
[58] S. A. Muhammad, Fraud: the affinity of classification
techniques to insurance fraud detection, Int J Innov Technol
Explor Eng 3 (11) (2014), 62–66.
[59] C. M. Fuller, D. P. Biros, and D. Delen, An investigation
of data and text mining methods for real world deception
detection, Expert Syst Appl 38 (7) (2011), 8392–8398.
[60] C.-H. Wen, P.-Y. Hsu, C-y Wang, and T. L. Wu, Identifying
smuggling vessels with artificial neural network and
logistics regression in criminal intelligence using vessels
smuggling case data, Intell Inform Database Syst Lect
Notes Comput Sci 7197 (2012), 539–548.
[61] M. Pandey, and V. Ravi, Detecting phishing e-mails
using text and data mining, In Proceedings of the IEEE
International Conference on Computational Intelligence &
Computing Research, Coimbatore, India, 2012, 18–20.
[62] R. P. Ang, and D. H. Goh, Predicting juvenile offending:
a comparison of data mining methods, Int J Offender Ther
Comp Criminol 57 (2) (2013), 191–207.
[63] N. Tollenaar, and P. G. M. van der Heijden, Which method
predicts recidivism best?: a comparison of statistical,
machine learning and data mining predictive models, J R
Stat Soc A 176 (2) (2013), 565–584.
[64] O. De-Vel, A. Anderson, M. Corney, and G. Mohay,
Mining e-mail content for author identification forensics,
ACM SIGMOD Rec 30 (4) (2001), 55–64.
[65] K. Kianmehr, and R. Alhajj, Effectiveness of support vector
machine for crime hot-spots prediction, J Appl Artif Intell
22 (5) (2008), 433–458.
[66] R. Abu Hana, C. Freitas, L. S. Oliveira, and F. Bor-
tolozzi,Crime Scene Classification. In Proceedings of the
23th Annual ACM Symposium in Applied Computing, 16-
20 March, Fortaleza, Cear, Brazil, 2008, 419–423.
[67] Z. Liu, D. Lin, and F. Guo, A method for locating
digital evidences with outlier detection using support vector
machine, Int J Netw Secur 6 (3) (2008), 301–308.
Statistical Analysis and Data Mining: The ASA Data Science Journal DOI:10.1002/sam
[68] R. Basili, C. Giannone, C. D. Vescovo, A. Moschitti,
and P. Naggar, Kernel-based relation extraction for crime
investigation. In AI*IA 2009: Emergent Perspectives in
Artificial Intelligence, 2009, 161–171.
[69] P. Wang, R. Mathieu, J. Ke, and H. J. Cai, Predicting crimi-
nal recidivism with support vector machine. In Proceedings
of the International Conference on Management and Ser-
vice Science, 24–26 August, Wuhan, 2010, 1–9.
[70] M. Salem, and S. Stolfo, Detecting masqueraders: a com-
parison of one-class bag-of-words user behavior model-
ing techniques, J Wireless Mob Netw Ubiquitous Comput
Dependable Appl 1 (1) (2010), 3–13.
[71] A. Modupe, O. O. Olugbara, and S. O. Ojo, Exploring
support vector machines and random forests to detect
advanced fee fraud activities on internet, In Proceedings
of the IEEE 11th Conference on Data Mining Workshops,
11 December, Vancouver, BC, 2011, 331–335.
[72] S. Bhattacharyya, S. Jha, K. Tharakunnel, and J. C.
Westland, Data mining for credit card fraud: a comparative
study, Decis Support Syst 50 (3) (2011), 602–613.
[73] H. M. Deylami, and Y. P. Singh, Adaboost and SVM based
cybercrime detection and prevention model, Artif Intell Res
1 (2) (2012), 117–130.
[74] R. Alwee, S. M. H. Shamsuddin, and R. Sallehuddin,
2013. Hybrid support vector regression and autoregressive
integrated moving average models improved by particle
swarm optimization for property crime states forecasting
with economic indicators, Sci World J, 1–11.
[75] R. Alwee, S. M. H. Shamsuddin, and R. Sallehuddin,
2013. Economic indicators selection for property crime
rates using Grey Relational Analysis and Support Vector
Regression. In Proceedings of the International Conference
on Systems, Control, Signal Processing and Informatics,
16–19 July, Rhodes Island, 178–185.
[76] A. K. Junoh, M. N. Mansor, A. M. Yaacob, F. A. Adnan,
S. A. Saad, and N. M. Yazid, Crime Detection with DCT
and artificial intelligent approach, Adv Mat Res (2013),
816–817. 610–615.
[77] J. Bendler, T. Brandt, S. Wagner, and D. Neumann, Inves-
tigating crime-to-twitter relationships in urban environ-
ments–facilitating a virtual neighborhood watch, In Pro-
ceedings of the 22nd European Conference on Information
Systems, Tel Aviv, 2014, 9–11.
[78] M. K. Sparrow, The application of network analysis to
criminal intelligence: an assessment of the prospects, Social
Netw 13 (3) (1991), 251–274.
[79] J. C. Wang, and C. Q. Chiu, Detecting online auction
inflated-reputation behaviors using social network analy-
sis, In Proceedings of the Annual Conference of the North
American Association for Computational Social and Orga-
nizational Science, Notre Dame, 2005, 26–28.
[80] J. Qin, J. J. Xu, D. Hu, M. Sageman, and H. Chen,
2005. Analyzing terrorist networks: a case study of the
global salafi jihad network. In Intelligence and Security
informatics, 3495, 287–304.
[81] D. McNally, and J. Alston, Use of social network analysis
(SNA) in the examination of an outlaw motorcycle gang,
J Gang Res 13 (3) (2006), 1–25.
[82] S. Ressler, Social network analysis as an approach to
combat terrorism: Past, present, and future research, Homel
Secur Aff 2 (2) (2006), 1–10.
[83] R. Rowe, G. Creamer, S. Hershkop, and S. J. Stolfo, Auto-
mated social hierarchy detection through email network
 Hassani et al.: Review of Data Mining Applications in Crime
analysis, In Proceedings of the 9th WebKDD and 1st SNA-
KDD 2007 Workshop on Web mining and Social Network
Analysis, San Jose, CA, 2007, 12–15.
[84] M. Chau, and J. Xu, Mining communities and their
relationships in blogs: a study of online hate groups, Int
J Hum Comput Stud 65 (1) (2007), 57–70.
[85] P. S. Nair, and S. T. Sarasamma, Data mining through
fuzzy social network analysis, In Proceedings of the Annual
Meeting of the North American, San Diego, CA, 2007,
24–27.
[86] Q. Liu, and C. Tang, S. Qiao, Q. Liu, F. Wen, 2007. Mining
the core member of terrorist crime group based on social
network analysis. Intelligence and Security Informatics,
4430, 311–313.
[87] J. C. Wang, and C. C. Chiu, Recommending trusted online
auction sellers using social network analysis, Expert Syst
Appl 34 (3) (2008), 1666–1679.
[88] F. Kerschbaum, and A. Schaad, Privacy-preserving social
network analysis for criminal investigations, In Proceed-
ings of the 7th ACM workshop on Privacy in the electronic
society, Alexandria, VA, 2008, 27–31.
[89] M. Chau, and J. Xu, Using web mining and social network
analysis to study the emergence of cyber communities in
blogs, Terror Inform 18 (2008), 473–494.
[90] S. Qiao, C. Tang, J. Peng, W. Liu, F. Wen, and J.
Qiu, Mining key members of crime networks based on
personality trait simulation email analysis system. Chin J
Comput 31 (10) (2008), 1795–1803.
[91] H. Chen, Homeland security data mining using social
network analysis. In Proceedings of the IEEE International
Conference on Intelligence and Security Informatics,
Taipei, xxvii, 2008, 17–20.
[92] A. M. Fard, and M. Ester, Collaborative mining in multiple
social networks data for criminal group discovery, In Pro-
ceedings of the International Conference on Computational
Science and Engineering, Vancouver, BC, 2009, 29–31.
[93] D. M. Schwartz, and T. D. A. Rouselle, Using social
network analysis to target criminal networks, Trends Organ
Crime 12 (2) (2009), 188–207.
[94] C. Chiu, Y. Ku, T. Lie, and Y. Chen, Internet auction fraud
detection using social network analysis and classification
tree approaches, Int J Electr Commer 15 (3) (2011),
123–147.
[95] P. J. Carrington, 2011. Crime and social network analysis.
In The SAGE Handbook of Social Network Analysis,
236–255.
[96] J. A. Johnson, and J. D. Reitzel, Geneva, 2011, Social
network analysis in an operational environment: defining
the utility of a network approach for crime analysis
using the Richmond City Police Department as a case
study. In: Proceedings of the International Police Executive
Symposium, 1–23.
[97] D. Prakash, and S. Suren, Detection and analysis of hidden
activities in social networks, Int J Comput Appl 77 (16)
(2013), 34–38.
[98] A. Borthwick, J. Sterling, E. Agichtein, and R. Grishman,
NYU: description of the MENE named entity system as
used in MUC-7, In Proceedings of the Seventh Message
Understanding Conference, 29 April - 1 May, Fairfax, VA,
1998.
[99] G. R. Krupka and K. Hausman, IsoQuest Inc.: description
of the NetOwl (TM) Extractor System as Used for MUC-
7, In Proceedings of the Seventh Message Understanding
Conference, 29 April–1 May, Fairfax, VA, 1998.
Statistical Analysis and Data Mining: The ASA Data Science Journal DOI:10.1002/sam
153
[100] I. H. Witten, Z. Bray, M. Mahoui, and W. J. Teahan,
Using language models for generic entity extraction,
In Proceedings of the 16th International Conference on
Machine Learning Workshop on Text Mining, Bled,
Slovenia, 1999.
[101] S. Baluja, V. O. Mittal, and R. Sukthankar, Apply-
ing machine learning for high-performance named-entity
extraction, Comput Intell 16 (4) (2000), 586–595.
[102] S. Miller, M. Crystal, H. Fox, L. Ramshaw, R. Schwartz,
R. Stone, and R. Weischedel, Algorithms that learn to
extract information BBN: description of the sift system as
used for MUC-7. In Proceedings of the Seventh Message
Understanding Conference, 29 April–1 May, Fairfax, VA,
1998.
[103] LexiQuest Mine. Executive Report, SPSS, IBM. URL
http://www.spss.ch/upload/1037287198_LexiQuestMine.
pdf. Accessed: January 26, 2016.
[104] A. Iriberri, and G. Leroy, Natural language processing and
e-government: extracting reusable crime report informa-
tion, In Proceedings of the IEEE International Conference
on Information Reuse and Integration, Las Vegas, IL, 2007,
13–15.
[105] J. J. Xu, and H. Chen, Fighting organized crimes: using
shortest-path algorithms to identify associations in criminal
networks, Decis Support Syst 38 (3) (2004), 473–487.
[106] W. R. Harper, and D. H. Harris, The application of link
analysis to police intelligence, Human Factors 17 (2)
(1975), 157–164.
[107] H. G. Goldberg, and R. W. H. Wong, Restructuring
transactional data for link analysis in the FinCEN AI
system, In AAAI Fall Symposia, Florida, Orlando, 1998,
22–24.
[108] P. Klerks, The network paradigm applied to criminal
organizations: theoretical nitpicking or a relevant doctrine
for investigators? Recent developments in the Netherlands,
Connections 24 (3) (2001), 53–65.
[109] T. Anderson, L. Arbetter, A. Benawides, and A. Longmore-
Etheridge, Security works, Security Manage 38 (17)
(1994), 17–20.
[110] R. Imieliski, T., and Swami, A. Agrawal, Mining Asso-
ciation Rules between sets of items in large databases, In
Proceedings of the 1993 ACM SIGMOD International Con-
ference on Management of data, 26–28 May, Washington,
DC, 1993, 207–216.
[111] R. Agrawal and R. Srikant, Fast algorithms for mining
association rules, In Proceedings of the 20th International
Conference on Very Large Data Bases, September 12–15,
Santiago de Chile, Chile, 1994, 487–499.
[112] S. Li, T. Wu, and W. M. Pottenger, Distributed higher order
association rule mining using information extracted from
textual data, SIGKDD Explorations 7 (1) (2005), 26–35.
[113] A. Asuncion, and D. J. Newman, Uci machine learning
repository. 2007, Irvine, CA, Department of Information
and Computer Science, University of California. URL
http://www.ics.uci.edu/mlearn/MLRepository.html.
[114] R. Berk, L. Sherman, G. Barnes, E. Kurtz, and L. Ahlman,
Forecasting murder within a population of probationers and
parolees: a high stakes application of statistical learning,
J R Stat Soc A, 172 (1) (2009), 971–1008.
[115] J. Cowie, and W. Lehnert, Information extraction, Commun
ACM 39 (1) (1996), 80–91.
[116] N. A. Chinchor, Overview of MUC-7/MET-2, In Proceed-
ings of the Seventh Message Understanding Conference,
29 April-1 May, Fairfax, VA, 1998.
154 Statistical Analysis and Data Mining: The ASA Data Science Journal, Vol. 9 (2016)
[117] H. Yun, D. Ha, B. Hwang, and K. H. Ryu, Mining
Association Rules on significant rare data using relative
support, J Syst Softw 67 (3) (2003), 181–191.
[118] X. Wu, V. Kumar, J. R. Quinlan, J. Ghosh, Q. Yang, H.
Motoda, G. F. McLachlan, A. Ng, B. Liu, P. S. Yu, Z.-H.
Zhou, M. Steinbach, D. J. Hand, and D. Steinberg, Top
10 algorithms in data mining, Knowl Inform Syst 14 (1)
(2008), 1–37.
[119] J. S. Park, M.-S. Chen, and P. S. Yu, An effective
hash-based algorithm for mining association rules, ACM
SIGMOD Record 24 (2) (1995), 175–186.
[120] J. S. Park, M. S. Chen, and P. S. Yu, Using a hash-based
method with transaction trimming for mining association
rules, IEEE Trans Knowl Data Eng 9 (5) (1997), 813–825.
[121] J. Han, and J. Pei, and Y. Yin, Mining frequent patterns
without candidate generation, ACM SIGMOD Record 29
(2) (1995), 175–186.
[122] J. Pei, J. Han, H. Lu, S. Nishio, S. Tang, and D. Yang,
H-mine: Hyper-structure mining of frequent patterns in
large databases, In Proceedings of the IEEE International
Conference on Data Mining, 29 November-2 December,
San Jose, CA, 2001, 441–448.
[123] B. Liu and W. Hsu, Integrating classification and associa-
tion rule mining. In: Proceedings of The 3rd International
Conference on Knowledge Discovery and Data Mining,
New York, 27-31 August, 1998.
[124] L. Breiman, J. Friedman, R. Olshen, and C. Stone,
Classification and regression trees, London, Chapman &
Hall/CRC, 1984.
[125] J. R. Quinlan, C4.5: Programs for Machine Learning,
Burlington, MA, Morgan Kaufmann, 1992.
[126] J. R. Quinlan, Induction of decision trees, Mach Learn
1(1) (1986), 81–106.
[127] J. W. Hunt, and T. G. Szymanski, A fast algorithm for com-
puting longest common subsequences, Communications of
the ACM 20(5) (1977), 350–353.
[128] M. Mehta, R. Agrawal, and J. Rissanen, SLIQ: a fast
scalable classifier for data mining. In Advances in Database
Technology EDBT’96, 1057 (1996), 18–32.
[129] J. C. Shafer, R. Agrawal, and M. Mehta, SPRINT:
a scalable parallel classifier for data mining, In Pro-
ceedings of the 22nd International Conference on Very
Large Databases, 3–6 September, Bombay, India, 1996,
544–555.
[130] C. Cortes, and V. Vapnik, Support vector networks, Mach
Learn 20 (3) (1995), 273–297.
[131] P. Langley, W. Iba, and K. Thompson, An analysis of
Bayesian classifiers, In Proceedings of the 10th National
Conference on Artificial Intelligence, 12–16 July, San Jose,
CA, 1992, 223–228.
Statistical Analysis and Data Mining: The ASA Data Science Journal DOI:10.1002/sam
[132] R. Kohavi, Scaling up the accuracy of naive-Bayes
classifiers: a decision-tree hybrid, In Proceedings of the
2nd International Conference on Knowledge Discovery and
Data Mining, Portland, OR, 1996, 2–4.
[133] H. Zhang, The optimality of naive Bayes, Proceedings
of the 17th International Florida Artificial Intelligence
Research Society Conference, Florida, Miami Beach, 2004,
12–14. May,
[134] M. D. Richard, and R. P. Lippmann, Neural network
classifiers estimate Bayesian a posteriori probabilities,
Neural Comput 3 (4) (1991), 461–483.
[135] G. P. Zhang, Neural networks for classification: a survey,
IEEE Trans Syst Man Cybern C Appl Rev 30 (4) (2000),
451–462.
[136] H. Gish, A probabilistic approach to the understanding and
training of neural network classifiers, In Proceedings of the
International Conference on Acoustics, Speech, and Signal
Processing, Albuquerque, NM, 1990, 3–6.
[137] P. A. Shoemaker, A note on least-squares learning
procedures and classification by neural network models,
IEEE Trans Neural Netw 2 (1) (1991), 158–160.
[138] E. A. Wan, Neural network classification: a Bayesian
interpretation, IEEE Trans Neural Netw 1 (4) (1989),
303–305.
[139] B. Widrow, D. E. Rumelhart, and M. A. Lehr, Neural
networks: applications in industry, business and science,
Commun ACM 37 (3) (1994), 93–105.
[140] J. Mena, Investigative data mining for security and criminal
detection, Oxford, Butterworth-Heinemann, 2003.
[141] A. M. Fard, and M. Ester, Collaborative mining in mul-
tiple social networks data for criminal group discovery,
In Proceedings of the International Conference on Com-
putational Science and Engineering, Vancouver, BC, 2009,
29–31.
[142] C. Haythornthwaite, Social network analysis: an approach
and technique for the study of information exchange, Libr
Inform Sci Res 18 (4) (1996), 323–342.
[143] R. D. Alba, Taking stock of network analysis: A decade’s
results, Res Sociol Organ 1 (1982), 39–74.
[144] J. Scott, Social Network Analysis: A Handbook, London,
SAGE, 1991.
[145] S. Wasserman, and K. Faust, Social Network Analysis
Methods and Applications, Cambridge, MA, Cambridge
University Press, 1995.
[146] K. Chan, and J. Liebowitz, The synergy of social network
analysis and knowledge mapping: a case study, Int J
Manage Decis Mak 7 (1) (2006), 19–35.
[147] K. Faust, Centrality in affiliation networks, Soc Netw 19
(2) (1997), 157–191.