Content uploaded by Wingyan Chung
Author content
All content in this area was uploaded by Wingyan Chung on Sep 15, 2013
Content may be subject to copyright.
Crime Data Mining: An Overview and Case Studies
Hsinchun Chen, Wingyan Chung, Yi Qin, Michael Chau, Jennifer Jie Xu, Gang Wang, Rong Zheng,
Homa Atabakhsh
{hchen, wchung, yiqin, mchau, jxu, gang, rong, homa}@bpa.arizona.edu
Artificial Intelligence Lab, Department of Management Information Systems, University of Arizona,
Tucson, AZ 85721, USA
http://ai.bpa.arizona.edu/
Abstract. The concern about national security has increased significantly since the 9/11 attacks.
However, information overload hinders the effective analysis of criminal and terrorist activities. Data
mining applied in the context of law enforcement and intelligence analysis holds the promise of
alleviating such problems. In this paper, we review crime data mining techniques and present four case
studies done in our ongoing COPLINK project.
1. Introduction
The concern about national security has increased significantly since the terrorist attacks on September
11, 2001. Intelligence agencies such as the CIA and FBI are actively collecting and analyzing information
to investigate terrorists’ activities. Local law enforcement agencies have also become more alert to
criminal activities in their own jurisdictions. One challenge to law enforcement and intelligence agencies
is the difficulty of analyzing large volumes of data involved in criminal and terrorist activities. Data
mining holds the promise of making it easy, convenient, and practical to explore very large databases for
organizations and users. In this paper, we review data mining techniques applied in the context of law
enforcement and intelligence analysis, and present four case studies done in our ongoing COPLINK
project (Hauck et al., 2002).
2. An Overview of Crime Data Mining
It is useful to review crime data mining in two dimensions: (1) crime types and security concerns and (2)
crime data mining approaches and techniques.
2.1 Crime Types and Security Concerns
Crime is defined as “an act or the commission of an act that is forbidden, or the omission of a duty that is
commanded by a public law and that makes the offender liable to punishment by that law” (Webster
Dictionary). An act of crime encompasses a wide range of activities, ranging from simple violation of
civic duties (e.g., illegal parking) to internationally organized crimes (e.g., the 9/11 attacks). Table 1
summarizes the different types of crimes in increasing degree of public influence. Note that both local and
national law enforcement and security agencies are facing many similar challenges.
2.2 Crime Data Mining Approaches and Techniques
Data mining is defined as the identification of interesting structure in data, where structure designates
patterns, statistical or predictive models of the data, and relationships among parts of the data (Fayyad &
Uthurusamy, 2002). Data mining in the context of crime and intelligence analysis for national security is
still a young field. The following describes our applications of different techniques in crime data mining.
Entity extraction has been used to automatically identify person, address, vehicle, narcotic drug, and
personal properties from police narrative reports (Chau et al., 2002). Clustering techniques such as
“concept space” have been used to automatically associate different objects (such as persons,
organizations, vehicles) in crime records (Hauck et al., 2002). Deviation detection has been applied in
fraud detection, network intrusion detection, and other crime analyses that involve tracing abnormal
activities. Classification has been used to detect email spamming and find authors who send out
unsolicited emails (de Vel et al., 2001). String comparator has been used to detect deceptive information
in criminal records (Wang et al., 2002). Social network analysis has been used to analyze criminals’ roles
and associations among entities in a criminal network.
Type
Local Law Enforcement Level
National Security Level
Traffic
Violations
Driving under influence (DUI), fatal/personal
injury/property damage traffic accident, road rage
-
Sex Crime
Sexual offenses, sexual assaults, child molesting
Organized prostitution
Theft
Robbery, burglary, larceny, motor vehicle theft,
stolen property
Theft of national secrets or weapon
information
Fraud
Forgery and counterfeiting, frauds, embezzlement,
identity deception
Transnational money laundering, identity
fraud, transnational financial fraud
Arson
Arson on buildings, apartments
-
Gang / drug
offenses
Narcotic drug offenses (sales or possession)
Transnational drug trafficking
Violent Crime
Criminal homicide, armed robbery, aggravated
assault, other assaults
Terrorism (bioterrorism, bombing,
hijacking, etc.)
Cyber Crime
Internet frauds, illegal trading, network intrusion/hacking, virus spreading, hate crimes,
cyber-piracy, cyber-pornography, cyber-terrorism, theft of confidential information
Table 1. Crime types at different levels
3. Case Studies of Crime Data Mining
Based on the crime characteristics and analysis techniques discussed above, we present four case studies
of crime data mining that are part of our ongoing COPLINK project.
3.1 Entity Extraction for Police Narrative Reports
Valuable criminal-justice data in free texts such as police narrative reports are currently difficult to be
accessed and used by intelligence investigators in crime analyses. We proposed a neural network-based
entity extractor, which applies named-entity extraction techniques to automatically identify useful entities
from police narrative reports of the Tucson Police Department (TPD). The system has three major
components: (1) Noun phrasing: It is a modified version of the Arizona Noun Phraser (Tolle & Chen,
2000) and extracts noun phrases as named entities from documents based on syntactical analysis; (2)
Finite state machine and lexical lookup: A finite state machine compares each word in the extracted
phrase, as well as the words immediately before and after the phrase, with the items in a handcrafted
lexicons. Each comparison will generate a binary value (either 0 or 1) to indicate a match or mismatch;
(3) Neural network: The feedforward/backpropagation neural network predicts the phrase’s most possible
entity type. Preliminary evaluation results demonstrated that our approach is feasible and has some
potential values for real-life applications. Our system achieved encouraging precision and recall rates for
person names and narcotic drugs (74 – 85%), but did not perform as well for addresses and personal
properties (47 – 60%) (Chau et al., 2002). Our future work includes conducting larger-scale evaluation
studies and enhancing the system to capture human knowledge interactively.
3.2 Detecting Criminal Identity Deceptions: An Algorithmic Approach
Criminals often provide police officers with deceptive identities to mislead police investigations, for
example, using aliases, fabricated birth dates or addresses, etc. The large amount of data also prevents
officers from examining inexact matches manually. Based on a case study on deceptive criminal identities
recorded in the TPD, we have built a taxonomy of criminal identity deceptions that consisted of name
deceptions, address deceptions, date-of-birth deceptions, and identity number deceptions. We found
criminals usually made minor changes to their real identity information. For example, one may give a
name similarly spelled or, change the sequence of digits in his social security number. Based on the
taxonomy, we developed an algorithmic approach to detect deceptive criminal identities automatically
(Wang et al., 2002). Our approach utilized four identity fields: name, address, date-of-birth, and social-
security-number and compared each corresponding field for a pair of criminal identity records. An overall
disagreement value between the two records was computed by calculating the Euclidean Distance of
disagreement measures over all attribute fields. A deception in this record pair will be noticed when the
overall disagreement value exceeds a pre-determined threshold value, which is acquired during training
processes. We conducted an experiment using a sample set of real criminal identity records from the TPD.
The results showed that our algorithm could accurately detect 94% of criminal identity deceptions.
3.3 Authorship Analysis in Cybercrime
The large amount of cyber space activities and their anonymous nature make cybercrime investigation
extremely difficult. Conventional ways to deal with this problem rely on a manual effort, which is largely
limited by the sheer amount of messages and constantly changing author IDs. We have proposed an
authorship analysis framework to automatically trace identities of cyber criminals through messages they
post on the Internet. Under this framework, three types of message features, including style markers,
structural features, and content-specific features, are extracted and inductive learning algorithms are used
to build feature-based models to identify authorship of illegal messages. To evaluate the effectiveness of
this framework, we conducted an experimental study on data sets of English and Chinese email and
online newsgroup messages produced by a small number of authors. We tested three inductive learning
algorithms: decision trees, backpropagation neural networks, and Support Vector Machines. Our
experiments demonstrated that with a set of carefully selected features and an effective learning
algorithm, we were able to identify the authors of Internet newsgroup and email messages with a
reasonably high accuracy. We achieved average prediction accuracies of 80% - 90% for email messages,
90% - 97% for the newsgroup messages, and 70% - 85% for Chinese Bulletin Board System (BBS)
messages. Significant performance improvement was observed when structural features were added on
top of style markers. SVM outperformed the other two classifiers on all occasions. The experimental
results indicated a promising future of using our framework to address the identity-tracing problem.
3.4 Criminal Network Analysis
In organized crimes such as narcotics trafficking, terrorism, gang-related crimes, and frauds, offenders
often cooperate and form networks to carry out various illegal activities. Social Network Analysis (SNA)
has been recognized as an appropriate methodology to uncover previously unknown structural patterns
from criminal networks. We have employed SNA techniques for criminal network analysis in our
COPLINK project. Four steps are involved in our analysis: (1) Network extraction: We utilized TPD
crime incident reports as sources for criminal relationship information because criminals who committed
crimes together usually were related. The concept space approach (Chen & Lynch, 1992) was used to
identify and uncover criminal relationships; (2) Subgroup detection: We employed hierarchical clustering
to detect subgroups in a criminal network based on relational strength; (3) Interaction pattern discovery:
We employed an SNA approach called blockmodeling to reveal patterns of between-group interaction.
Given a partitioned network, blockmodel analysis determines the presence or absence of an interaction
between a pair of subgroups by comparing the density of the links between these two subgroups to a
predefined threshold value; (4) Central member identification: We employed several measures, such as
degree, betweenness, and closeness to identify central members in a given subgroup. These three
measures can suggest the centrality of a network member.
Figure 1 shows a narcotics network consisting of 60 criminals. It is difficult to detect subgroups,
interaction patterns, and the overall structure from this original network manually. Using clustering and
blockmodeling methods, however, a chain structure became apparent (Figure 1b). We have conducted a
field study at the TPD with three domain experts who confirmed that the subgroups and central members
found by the system were correct representations of the reality. They believed that this system would be
very useful for crime investigation and could greatly increase crime analysts’ productivity.
(a)
(b)
Figure 1(a). A 60-member narcotics network. (b) The chain structure (thicker links) found.
4. Conclusions and Future Directions
In this paper, we have presented an overview of crime data mining and four COPLINK case studies. From
the encouraging results, we believe that crime data mining has a promising future for increasing the
effectiveness and efficiency of criminal and intelligence analysis. Many future directions can be explored
in this still young field. For example, more visual and intuitive criminal and intelligence investigation
techniques can be developed for crime pattern and network visualization.
5. Acknowledgements
This project has primarily been funded by NSF Digital Government Program, “COPLINK Center:
Information and Knowledge Management for Law Enforcement,” #9983304, July 2000 – June 2003. We
would like to thank the following people for their support and assistance during the entire project
development and evaluation process: All members of the University of Arizona Artificial Intelligence Lab
staff and specifically past and present COPLINK team members; Lt. Jenny Schroeder, Detective Tim
Petersen and other contributing personnel from the Tucson Police Department; and other contributing
members of the Phoenix Police Department.
6. References
Chau, M., Xu, J., & Chen, H. (2002). Extracting meaningful entities from police narrative reports. In:
Proceedings of the National Conference for Digital Government Research (dg.o 2002), Los Angeles,
California, USA.
Chen, H., & Lynch, K.J. (1992). Automatic construction of networks of concepts characterizing document
databases. IEEE Transactions on Systems, Man, and Cybernetics, 22(5), 885-902.
de Vel, O., Anderson, A., Corney, M., & Mohay, G. (2001). Mining E-mail Content for Author
Identification Forensics. SIGMOD Record, 30(4), 55-64.
Fayyad, U.M., & Uthurusamy, R. (2002). Evolving data mining into solutions for insights.
Communications of the ACM, 45(8), 28-31.
Hauck, R.V., Atabakhsh, H., Ongvasith, P., Gupta, H., & Chen, H. (2002). Using Coplink to analyze
criminal-justice data. IEEE Computer, 35(3), 30-37.
Tolle, K.M., & Chen, H. (2000). Comparing noun phrasing techniques for use with medical digital library
tools. Journal of the American Society for Information Science, 51(4), 352-370.
Wang, G., Chen, H., & Atabakhsh, H. Automatically detecting deceptive criminal identities.
Communications of the ACM (Accepted for publication, forthcoming).
