You are on page 1of 5

An Analysis of Data Mining Applications

in Crime Domain

P. Thongtae and S. Srisuk

The applications in law enforcement from data analysis


Abstract— In this paper, we give comprehensive surveys of can be categorized into two critical elements, crime control
efficient and effective methods/techniques on data mining for and criminal suppression. Crime control tends to use
crime data analysis. These techniques aim at finding the illegal knowledge from the analyzed data to control and prevent the
activities of professional identity fraudsters based on knowledge
discovered from their own histories. We also raise some
occurrence of crime, while the criminal suppression tries to
problems of applied data mining in crime control and criminal catch a criminal by using his/her history recorded in data
suppression. It is known that detecting crime from data analysis mining.
can be difficult because daily activities of criminal generate Brown [2] has constructed a software framework for
large amounts of data and stem from various formats. In mining data in order to catch professional criminals. They
addition, the quality of data analysis depends greatly on proposed an information system that can be used to catch the
background knowledge of analyst. A criminal can range from
civil infractions such as illegal driving to terrorism mass
criminals in their own regions. The software can be used to
murder such as the 9/11 attacks, therefore it is difficult to model turn data into useful information with two technologies, data
the perfect algorithm to cover all of them. Finally, this paper fusion and data mining. Data fusion manages, fuses and
proposes a guideline to overcome the problem. interprets information from multiple sources, and it
overcomes confusion from conflicting reports and cluttered
Index Terms—Crime data, Crime data mining, Crime or noisy backgrounds. Data mining is concerned with the
control, Criminal suppression automatic discovery of patterns and relationships in large
databases. His software was called ReCAP (Regional Crime
Analysis Program), which was built to provide crime analysis
I. INTRODUCTION
with both technologies.
Data mining is a powerful technique with great potential When the terrorism was exploded by 9/11 attacks, concern
to help criminal investigators focus on the most important about national security has increased significantly and the
information in their crime data. The knowledge discovered world has changed forever. The strategic against a terrorist
from existing data leads to reveal a value-added of its must be more advanced in order to prevent suicide attacks
information. Successful data mining methodology depends from their stratagem. In the congressional conference, Robert
heavily on the particular choice of techniques used by S. Mueller – The director of investigative department of FBI,
analyst. Their practical applications are, for example, the indicated that the excessively emphasize to catch the criminal
criminal investigation, market sale prediction and playing with slightly attention for crime control is the main problem
behavior analysis in sport games. However the more data and of law enforcement in the world [4]. It is more interesting
more complex queries being processed and maintained, the now in data gathering for criminal control plan.
more powerful the system is required. This includes the Abraham et. al. [1] proposed a method to employ computer
capability of the system to analyze large amount of data and log files as history data to search some relationships by using
complex information from diverse sources the frequency occurrence of incidents. Then, they analyze the
In crime control of the law enforcement, there are many result to produce profiles. The profiles could be used to
storage data and formats that have to be discovered. When its perceive the behavior of criminal. It should be noted that the
quantity has grown, it is difficult to analyze and search some types of crime could be changed over time influenced by the
new knowledge from them. Therefore, the data mining is variations of globalization and technology. Therefore, if we
applied to crime control and criminal suppression by using want to prevent the crime efficiently, the behavior must be
frequency occurrence and distance method under which these used with another kind of knowledge. We need to know the
assumptions can be accomplished. All these techniques give crime trends. de Bruin et. al. [3] introduced a new distance
results to benefit investigators in searching behavioral measure for comparing all individuals based on their profiles
patterns of professional criminal. and then clustering them accordingly. This method yields a
visual clustering of criminal careers and enables the
identification of classes of criminals. They demonstrated the
P. Thongtae is with the Department of Computer Engineering, applicability of data mining in the field of criminal career
Mahanakorn University of Technology, Bangkok THAILAND. (e-mail: analysis. Four important factors play a role in the analysis of
securitythai@gmail.com).
S. Srisuk is with the Department of Computer Engineering, Mahanakorn criminal careers: crime nature, frequency, duration and
University of Technology, Bangkok THAILAND. (e-mail: severity. They also develop a specific distance measure to
sanun@mut.ac.th).
combine this profile difference with crime frequency and B. Data Mining Processes
change of criminal behavior over time. The matrix was 1. Data Preprocessing
created that describes the amount of variation in criminal A data preprocessing is a data gathering, which is collected
careers between all couples of perpetrators. The data analysis from various sources. The preprocessing process consists of
can be used to determine the trend of criminal careers. Nath data cleaning, data integration and data transformation which
[6] proposed a method for data partitioning in order to use is usually processed by a computer program. It intends to
them to present in the form of geographic map. We could reduce some noises, incomplete and inconsistent data. The
decide the data partitions to be the trend of offend across results from preprocessing step can be later proceeding by
many areas. data mining algorithm. Some approaches of data
The organization of this paper is as follows. Section II preprocessing are regression and sampling.
describes a literature review on popular techniques that are 2. Data mining primitive
used in data mining for crime analysis. Relevant research is A data mining primitive defines data mining tasks as what
presented in section III. We give a discussion in section IV. we perform in data mining process. It depends on a
Finally, section V concludes the research issues in data hypothesis of analyst and how we want to prove that
mining. hypothesis. It is likely a guideline that we must choose before
we actually do the data mining process. The data mining
primitive consists of many procedures as follows:
II. LITERATURE REVIEW 2.1 Task-relevant data: It defines which data we are
interested in their dimensions and attributes. It also defines a
A. Popular Techniques in the Data Mining
data grouping criteria for data aggregation, such as data cube.
Several techniques have been proposed for solving a 2.2 Type of knowledge: It defines data mining techniques
problem of extracting knowledge from explosive data, each that are best suited for a problem at hand. The types of
of which adopts different algorithm. One of the areas where knowledge are characterization, discrimination, association,
information plays an important role is that of law classification/prediction, clustering and outlier-analysis.
enforcement. Obviously, the growing amount of criminal 2.3 Background knowledge: It is a predefined rule set for
data gives rise to various problems including data storage, making a decision in data mining. Examples of which is size
data warehousing and data analysis. The main ideas, used in of street is smaller than size of city and the age between 20
data mining, can be categorized as follows [4]: and 39 is identified as young. This can be defined by using a
1. Association Mining common rule set or fuzzy set indicating a level of condition.
A main idea of association mining technique is to search a 2.4 Measurements of pattern interestingness: A
relationship of attributes and tuples, by discovering measurement of pattern interestingness is criterions to
frequently occurring item sets in database. A result is patterns determine a level of interest. There are many factors that are
described as rules that represent one-way relationship. used in measurements including simplicity, certainty, utility
Furthermore, result rules consist of a confidential value and and novelty. In addition, an evaluation in the level of interest
support value, a value of which is used to identify the pattern. depends heavily on knowledge of researchers. For instance,
The support is a number of instances that complies with the if relation rule is too long and the level of interest (support
rules, whereas the confidential is a percentage of instances value) is low, then researcher may ignore this rule because it
that must likewise be complied by rules. In basket analysis, is low interesting.
for example, the association mining is a customer’s behavior 2.5 Visualization of discovered patterns: Visualization of
analysis that determines the products the customer frequently discovered pattern is a form of result from data mining in
buy together. which the outcome is being display. There are various forms
2. Classification and Prediction that would be appropriate for the knowledge background of
The classification and prediction approaches are based on researcher. The examples of visualizations are relation rules,
a divide-and-conquer technique by which the data are graphs, pie/bar chart and tables.
grouped and missing values are predicted. The data, with As mentioned before, the data mining primitive must be
sharing common properties among different crime entities, proceed before we actually start to perform data mining
are specified into predefined classes. A main target process. Some tasks are normally related. For example, if we
application is credit card approval, target marketing and choose the type of knowledge to be “association”, we also
medical diagnosis. determine in the length of rules and “support value” as
3. Clustering measurements of pattern interestingness. The visualization is
Clustering is a method to group data into classes with the relation rules.
identical characteristics in which the similarity of intra-class 3. Data mining
is maximized or minimized. An investigator may use this The final step is the data mining process which is used to
technique to identify suspects who conduct crimes in similar extract hidden predictive information from large crime
ways or discriminate among groups belonging to different database. Data mining tools predict future trends and
gangs. A distance measure is often used as a data partitioning behaviors of criminals, allowing investigators to make
function. However, in clustering technique data are usually knowledge-driven decisions. It can be implemented rapidly
not label in advance. on software and hardware platforms to enhance the value of
existing information. When data mining is developed in a
high performance computer or a complex grid computing, it
can be used to analyze massive crime data to deliver answer rather than a redundant one. Normally, an analysis in
to investigators such as "Where is the highest probability of geo-graphic data is harder than a query process in database
crime occurrences in Friday?" The result is then evaluated to system, such as an attributed matching based on distance
determine the value of knowledge. measure is computed as line of sight, travel distance and its
shortest path. Hence, its computation is much more complex
III. RELEVANT RESEARCH WORK than querying in usual database system. In addition, the
results from geo-graphic data mining can be applied
A. The Regional Crime Analysis Program immediately in crime analysis to identify when and where the
crime occurs. Spatial data mining from GIS and DBMS
In 1998, the USA law enforcement increasingly required a systems enables investigators who may lack extensive
database management system and a geographic information training as data analysts to explore a large crime database
system to support their crime analysis. Newly crime analysts with a useful geometric data for law enforcement.
manually used these systems to search an interesting data.
ReCAP Data Flow Diagram
They then analyzed data in order to discover its relationship.
Finally, the result from data analysis could be used to display
on the geographic map [2]. Routine Reporting
Pistol 2000 Analysis Database Visual Basic
A great experiment of laborious research on data mining Visual FoxPro Visual FoxPro

with crime data analysis has been explored in the same year.
Charlottesville City and Albemarle County areas were a
relatively safety community within a central Virginia. The
number of reported criminal incidents in this area was Geographic
Database Clustering
Information System Control Charting
typically 30,000 events per year. In contrast, Richmond City Querying MapInfo Visual Basic
Algorithms
Visual Basic
Visual Basic Visual Basic
alone has approximately 100,000 events of incidents per
year. The analyst needed to get the details of data such as
where incidents occurred and when. They needed to query
incidents from database management system in order to
generate all reports. Then, the results queried from database
Printed Results
are turned into useful information to display on the Disk Storage
Graphical User Interface

geographic map and to determine the pattern of relationships. Visual Basic

It should be noted that, it took three minutes to search and Figure 1: A diagram of ReCAP [2]
analyze and took five queries per report. Therefore, the
analyst might spend 450,000 minutes for Charlottesville City
and 1.5 million minutes for Richmond City. Clearly, It should be emphasized that this is not a new approach in
manually analysis of crime data is infeasible. data mining. It is only a feasible framework in development
The data mining is therefore a must choice that can be process. However, one attractive feature of the approach is
used to support a geographic information system. It reduces that knowledge from data analysis can be presented in a form
data analysis time consuming and enlarges a frontier of crime of geo-graphic map.
analysis. The knowledge from crime analysis can be
available by which the data mining is used as a core process B. Investigative Profiling
in data analysis.
Brown [2] proposed a framework for regional crime Investigative profiling can be narrowly used to discover
analysis program (ReCAP). The data mining was adopted as one or many criminals. It can also be used to search
an algorithm for crime data analysis. The core concept of knowledge from a large amount of data. Normally,
ReCAP program consists of: association rules are frequently used to search customer
4.1.1 Meta-Data: A data context described about data itself personalization, by providing a relationship between items.
such as pistol data and periodic report. In a marketing research, this technique can be used to search
4.1.2 Process and Display: An analysis process of data pleasure and behavior of consumers, i.e., their buying
such as database querying and clustering algorithm. A pattern. Therefore, retailers can conclude from relationship
displaying system is a graphical information system that is and offer products or services to individual, and lastly a
used to turn data into useful spatial information. personal profile can be created from customer buying pattern.
In figure 1, the system is a general data mining, its In 2002, a profiling technique has been applied in order to
structure is aimed to be used in geo-graphic data mining. In solve a problem in an investigation analysis. They interpreted
addition, the author provides a software framework for the results from data mining as an association rule sets which
spatial data mining. This method employs two significant is created from log data. The researcher then analyzed them
technologies, GIS and DBMS, to support a broader range of to create the perpetrator profiles.
crime analytical function. Information queried from Typically, the perpetrator profiling can be categorized into
geo-graphic data consume a long period of time. Therefore, if two types, actual and behavior profiles. Example of actual
we can construct a technique to filter data such as eliminating profile name is status of hiring and username, and of behavior
a small side road that we could not see, the analyst would profile is log file transactions, header and body of electronic
concentrate on searching relationship of more important data mail and call recording. By creating profiling by using
Crime Severity Difference
association rules, the results are the rules of relationship in
Minor Intermediate Severe
the same or different profile type. For example, a rule # Fac. # Fac. # Fac. Summed
Ind. 1 0.1 1 0.5 2 0.4 3
(StaffType = admin) ^ (DayofWeek = tuesday) ^ 0.1 1.0 1.2 2.3
(Application = database) ⇒ (Access = valid)
Ind. 2 0.3 1 0.4 2 0.3 3
0.3 0.8 0.9 2.0
Diff. 0.3
is a relationship between actual profile (admin) and behavior Total Profile Difference: 0.9
profile (access time). The meaning of this rule is that “the
administrator has authorization to access the database Table 2: The calculation for difference of crime nature and
program.” severity [3]
In an analysis process, log files can be obtained from
various computers and network devices such as system log Therefore, if the criminal profiles are analyzed for each
files, database log files and personal user files. The log files year, the trends of criminal career and crime varying would
are then processed by using association mining rules, the be revealed. The results from this knowledge discovery can
results present in the rules of profile relationship. However, be used to define the priority in which what types of victim
the rules are the item set that usually created with support must be protected. However, this method is time-consuming,
value. The remaining rules are pruned. Additional technique it is therefore not appropriate to use this algorithm with an
is used as “contrasting raw data to profiles” in order to enormous data.
change unexpected information to be profiles again. Hence, it
is possible to represent profiles as abnormal behavior of user. D. Crime Pattern Detection Using Data Mining
Accordingly, investigative profiling by the association
mining rules can be used to discover the pattern of crime that Nath [6] outlined a new approach for crime pattern
is frequently occurred. It can also reveal the behavior that is discovery. It is rather complex because the system searches
hidden or abnormal. By using this technique, the factor of knowledge from a huge data. Each record is composed of
crimes can be clearly illustrated such as environments, many attributes which describes the crimes, such as
properties of victims and criminal types. In particular, date/time, location, outline, demographic and weapon.
knowledge from investigative profiling leads to a plan for However, only a small portion of which relates to illegal
preventing and destroying opportunities of crime. activities which is useful for crime pattern construction.
However, in this method the relevance of attributes must Clustering techniques group entities into classes with
be selected by analyst before actually mining process. similar characteristics, i.e. data in similar characteristic is
Therefore, the results of data mining also depend on the defined to be the same class. Each group is then shown on
background knowledge of the analyst. geo-spatial information which describes distribution of crime
with different colors.
C. Data Mining Approaches to Criminal Career Analysis The results from a clustering technique can be used to
identify escape criminal. Incidents of the same group may
The criminal career analysis is an analyzing of data in have some relationships rise from suspects who conduct
order to determine a trend changing in offended behavior. In crimes in similar ways. Therefore, evidences from the
2006, the interested attributes were extracted from criminal incidents can be proved that it is collected from the same
profiles, example of which is crime nature, frequency, group. For instance, injured person in the incident said that
duration and severity. The interested attributes is then the offender was black hair, a witness said that the offender
processed by a clustering technique. Firstly, this technique was middle age, and injured person confirmed that he saw a
determines percentage of severity in each crime type as tattoo on the offender’s left arm.
shown in Table 1. Hence, the law enforcement can take this technique to
Crime type Traffic Financial … Sex analyze the pattern of crime data, by which the knowledge is
Crimes Crimes Crimes used to catch the escape offender.
Severity minor minor … severe
Percentage 0.0 0.6 … 0.1
Table 1: A criminal’s profile [3]
IV. DISCUSSION
Next, the algorithm calculates the difference in each We can conclude from a comprehensive survey that most
profile class such as crime nature and severity (Table 2). It of works analyze large amount of crime data. The algorithm
then sums all of difference which is computed with a number in meta-data and data mining techniques will be selected
of crimes for one year. The result is a difference of criminal based on background knowledge of analyst. Therefore, it is
career as two separated data groups, the first one is a general possible to choose inappropriate data items which turn into
street crime the other is a serious crime. redundant work when they must repeatedly performing
experiments. As a result, knowledge from crime analysis in
Crime Nature Difference data mining depends heavily on the background knowledge
Ind. 1 0.1 0.0 0.0 0.0 0.5 0.4 0.0 0.0 of analysts.
Ind. 2 0.0 0.2 0.1 0.0 0.4 0.3 0.0 0.0 summed
Diff. 0.1 0.2 0.1 0.0 0.0 0.0 0.0 0.0 0.6
However, most of analysts may have a computer skill but
they may lack extensive training on discovering the key
factors in crime analysis. Moreover, they can not make sure
that the result from crime data mining is best suited for real
situation in criminal investigations. Finally, they may do not
know that how to use it in practice.
We believe that the statistical method can be used to solve
the addressed problem. There are two interesting issues that
the statistical method can be applied in the future research
work. The first one is an attribute selection in which time
consuming can be reduced. It also helps to decide which data
mining technique can be adopted. The second one is the
quality assessment in the result of the crime data mining. It
can be used in appropriate situation so that a more quality
better could be obtained.

V. CONCLUSION
Data mining has been as active research area for a last
decade. We have shown its usefulness to a variety of tasks
including crime control and criminal suppression. It should
be pointed out that the result from data mining process is
useful knowledge that may be changed over time. In a
conventional system, they started by proposing a design and
development processes of information system for catching
criminal, so the data mining was just a database system based
SQL query command. When we were entered to a terrorism
era, the data mining has become to be knowledge discovery
for crime pattern understanding. Several data mining
techniques have been adopted in order to conquer with
terrorist. For example, the mining association rules can be
used to create the offender’s profile. The clustering technique
composes of two subsequent methods. The first one is the
partitioning which is a method to compute a difference of
crime career. We can also perceive the trends of crime pattern
that changes over year. The second one is data classification
in which the data in similar characteristics is classified as the
same group. Therefore, it can be used to identify the escape
offender. It is very encouraging finding that the crime pattern
understanding can be discovered from data analysis. Finally,
it is also possible to construct a novel method where the
quality of research is conducted from crime prediction point
of view.

REFERENCES
[1] T. Abraham and O. de Vel, "Investigative profiling with computer
forensic log data and association rules," in Proceedings of the IEEE
International Conference on Data Mining (ICDM'02), pp. 11 – 18,
2006.
[2] D.E. Brown, "The regional crime analysis program (RECAP): A
frame work for mining data to catch criminals," in Proceedings of the
IEEE International Conference on Systems, Man, and Cybernetics,
Vol. 3, pp. 2848-2853, 1998.
[3] J.S. de Bruin, T.K. Cocx, W.A. Kosters, J. Laros and J.N. Kok, “Data
mining approaches to criminal career analysis,” in Proceedings of the
Sixth International Conference on Data Mining (ICDM’06), pp.
171-177, 2006.
[4] J. Han and M. Kamber, “Data Mining: Concepts and Techniques,”
Morgan Kaufmann publications, pp. 1-39, 2006.
[5] J. Mena, “Investigative Data Mining for Security and Criminal
Detection”, Butterworth Heinemann Press, pp. 15-16, 2003.
[6] S.V. Nath, “Crime pattern detection using data mining,” in
Proceedings of the 2006 IEEE/WIC/ACM International Conference
on Web Intelligence and Intelligent Agent Technology, pp. 41-44,
2006.

You might also like