Crime Reduction Policy

ISSN: 2229-6959 (ONLINE) ICTACT JOURNAL ON SOFT COMPUTING, APRIL 2017, VOLUME: 07, ISSUE: 03 DOI: 10.21917/ijsc.2017.0202

1459

SURVEY ON CRIME ANALYSIS AND PREDICTION USING DATA MINING

TECHNIQUES

H. Benjamin Fredrick David1 and A. Suruliandi2 Department of Computer Science and Engineering, Manonmaniam Sundaranar University, India

Abstract

Data Mining is the procedure which includes evaluating and

examining large pre-existing databases in order to generate new

information which may be essential to the organization. The extraction

of new information is predicted using the existing datasets. Many

approaches for analysis and prediction in data mining had been

performed. But, many few efforts has made in the criminology field.

Many few have taken efforts for comparing the information all these

approaches produce. The police stations and other similar criminal

justice agencies hold many large databases of information which can

be used to predict or analyze the criminal movements and criminal

activity involvement in the society. The criminals can also be predicted

based on the crime data. The main aim of this work is to perform a

survey on the supervised learning and unsupervised learning

techniques that has been applied towards criminal identification. This

paper presents the survey on the Crime analysis and crime prediction

using several Data Mining techniques.

Keywords: Criminology, Crime Analysis, Crime Prediction, Data Mining

1. INTRODUCTION

Historically solving crimes has been the right of the criminal

justice and law enforcement specialists. With the increase in the

use of the computerized systems to track crimes and trace

criminals, computer data analysts have started lending their hands

in helping the law enforcement officers and detectives to speed up

the process of solving crimes. Criminology is process that is used

to identify crime and criminal characteristics. The criminals and

the crime occurrence possibility can be assessed with the help of

criminology techniques. The criminology aids the police

department, the detective agencies and crime branches in

identifying the true characteristics of a criminal. The criminology

department has been used in the proceedings of crime tracking

ever since 1800. Crimes are a social nuisance and cost our society

dearly in several ways. Even, the Indian Government has taken

steps to develop applications and software for the use of State and

Central Police in relation with the National Crime Records Bureau

(NCRB) [27]. Any research that can help in solving crimes faster

will pay for itself. About 10% of the criminals commit about 50%

of the crimes [15]. People who study criminology will be able to

identify the criminals based on the traces, characteristics and

methods of crime which can be collected from the crime scene. In

the middle of 1990s, data mining came into existence as a strong

tool to extract useful information from large datasets and find the

relationship between the attributes of the data [11]. Data mining

originally came from statistics and machine learning as an

interdisciplinary field, but then it was grown a lot that in 2001 it

was considered as one of the top 10 leading technologies which

will change the world [12]. According to many researchers such

as Nath [23], solving crimes is a difficult and time consuming task

that requires human intelligence and experience and data mining

is one technique that can help us with crime detection problems.

For solving crimes faster we have to develop a data mining

paradigm that performs an interdisciplinary approach between

computer science and criminal justice. As said earlier, the

Criminology is a process that aims to identify crime

characteristics and it is one of the most important fields for

applying data mining. By using this, data mining algorithms will

be able to produce crime reports and help in the identification of

criminals much faster than any human could. Because of this

remarkable feature, there is a growing demand for data mining in

criminology. Actually, Crime analysis is a process which includes

exploring the behavior of the crimes, detecting crimes and their

relationships with criminals. The huge volume of crime and

criminal datasets and the complexity of relationships between

these kinds of information have made criminology an appropriate

field for applying data mining techniques. Identifying crime

characteristics is the first step for proceeding with any further

analysis. The quality of data analysis depends greatly on

background knowledge of analyst. A criminal can range from

civil infractions such as illegal driving to terrorism mass murder

such as the 9/11 attacks, therefore it is difficult to model the

perfect algorithm to cover all of them [21]. The knowledge that is

gained from Data Mining approaches is a very useful and this can

help and support, the police. More specifically, we can use

classification and clustering based models to help in identification

of crime patterns and criminals. The wide range of data mining

applications in the criminology has made it an important field of

research. Data mining systems have played as a key role in

assisting humans in this forensic domain and criminology domain.

This makes it one of the most challenging decision-making environments for research.

The motivation for proceeding with this survey work is to aid

a helping hand to the young researchers who are performing their

research in criminal analysis and crime prediction areas. The

paper is organized in such a manner to provide insights about the

crime analysis procedure and then produce different types of

crime analysis operations and those which can be applied together

for producing an end user product which can be applied to the

crime analysis in any police stations and detective agencies. This

work will be a valuable reference to those who precede their

research work in the crime analysis and Crime prediction using

data mining techniques.

This survey paper is organized in such a manner for easy

understanding of the concepts. The general crime analysis

procedure is discussed in section 2. The Criminal analysis

methods are discussed in the section 3 which will include all the

different types of methods grouped under their own categories.

Finally, section 4 gives the Qualitative analysis of the Crime

Analysis and Prediction techniques and section 5 gives the

H. BENJAMIN FREDRICK DAVID AND A. SURULIANDI: SURVEY ON CRIME ANALYSIS AND PREDICTION USING DATA MINING TECHNIQUES

1460

Quantitative analysis of the Crime Analysis and Prediction

techniques.

2. CRIME ANALYSIS PROCEDURE

Usually, the crime analysis tasks can be a tedious process for

the police or the investigation team to work with. The criminals

when leaving the crime scene does leave some traces which can

be used as a clue to identify the criminals. The crime sequence

and the patterns which several criminals follow when committing

a crime make it easy for analyzing the crime. This process

includes several procedures to be followed in order to identify the

criminals and getting more information based only on the clues or

information given by the local people. The criminal can be

analyzed based on the information from the crime scene which is

tested against the previous crime patterns and judging by the

method which is implied to test and proceed with the information

that can affect the prediction results. The prediction can be further

made useful for detecting the crimes in advance or by adding more

cops to the sensitive areas which are identified by the system .The

police stations can put up special force when there are chances for

crime ahead of time. This type of the system will ensure there are

peace and prosperity among the citizens.

The crime analysis can be performed procedure which is similar

to figure Fig.1 which specifies each module which is used for

machine learning to predict the crime or form group of clusters of

criminals according to crime records. The criminals can hold

certain properties and their crime characteristics and crime careers

may vary from one criminal to another. Such a type of information

can be taken as the input dataset. The input dataset is given to a pre-

processor which performs the preprocessing based on the

requirements. Once the pre processing is completed the features or

attributes from those information are extracted which may be in the

form of text content from emails, the crime factors for a day,

criminal characteristics, geo-location of the criminal, etc,. The pre

processed result is further given to the classification algorithm or

the clustering algorithm based on the requirements. The

requirements may be anything from selecting the crime prone areas

to predicting the criminal based on the previous crime records.

Fig.1. Crime Prediction and Crime clustering based on the input

dataset

The classification algorithm works in a supervised learning

manner in which the training and testing phase is required in order

to train the classifier to identify the new unknown crime record.

This is known as prediction. Whereas the clustering algorithm

works in an un-supervised learning manner which automatically

separates the crime records based on the number of groups to be

created. The groups created in such a manner are known as

clusters. Such a type of design can be a general template for

applying crime prediction and crime analysis based on data

mining algorithms.

3. CRIMINAL ANALYSIS METHODS

3.1 TEXT, CONTENT AND NLP-BASED METHODS

Sharma [1] proposed a concept which depicts zero crime in

the society. For detecting the suspicious criminal activities, he has

concentrated on the importance of data mining technology and

designed a proactive application for that purpose. In his paper, he

proposed a tool which applies an enhanced Decision Tree

Algorithm to detect the suspicious e-mails about the criminal

activities. An improved ID3 Algorithm with an enhanced feature

selection method and attribute-importance factor is applied to

produce a better and faster Decision Tree based on the

information entropy which is explicitly derived from a series of

training data sets from several classes. He proposed a new

algorithm which is a combination of Advanced ID3 classification

algorithm and enhanced feature selection method for the better

efficiency of the algorithm.

Hamdy et al. [8] described an approach based on the people’s

interaction with social networks and mobile usage such as

location markers and call logs. Their work also introduced a

model for detecting suspicious behavior based on social network

feeds and it not only describes a new method using the social

interaction of people but, their work proposes a new system to

help crime analysis create faster and precise decisions. The

suspicious movement of the entity can be determined using the

sequence of inference rules. Their constructed model is able to

predict and characterize human behavior from reality data sources

3.2 CRIME PATTERNS AND EVIDENCE-BASED METHODS

Bogahawatte and Adikari [2] proposed an approach in which

they highlighted the usage of data mining techniques, clustering

and classification for effective investigation of crimes and criminal

identification by developing a system named Intelligent Crime

Investigation System (ICSIS) that could identify a criminal based

up on the evidence collected from the crime location. They used

clustering to identify the crime patterns which are used to commit

crimes knowing the fact that each crime has certain patterns. The

database is trained with a supervised learning algorithm, Naïve

Bayes to predict possible suspects from the criminal records. His

approach includes developing a multi-agent for crime pattern

identification. There are agents for the place, time, role trademark

and substance of criminals which separates the role of the

criminals in components. The system is a multi-agent system and

made with managed Java Beans. It makes it easy to encapsulate

the requested entities in the work into objects and returns it to the

bean for exposing properties. Classifying the criminals/ suspects is

based on the Naïve Bayes classifier for identifying most possible

suspects from crime data. Clustering the criminals is based on the

model to help to identify patterns of committing crimes.

Agarwal et al. [3] used the rapid miner tool for analyzing the

crime rates and anticipation of crime rate using different data

Input dataset

Preprocessing

Feature Selection

Classification Clustering

Crime Prediction Crime clusters

ISSN: 2229-6959 (ONLINE) ICTACT JOURNAL ON SOFT COMPUTING, APRIL 2017, VOLUME: 07, ISSUE: 03

1461

mining techniques. Their work done is for crime analysis using

the K-Means Clustering algorithm. The main objective of their

crime analysis work is to extract the crime patterns, predict the

crime based on the spatial distribution of existing data and

detection of crime. Their analysis includes the tracking homicide

crime rates from one year to the next

Kiani et al. [4] performed a crime analysis work based on the

clustering and classification techniques. Their work includes the

extraction of crime patterns by crime analysis based on available

criminal information, prediction of crimes based on the spatial

distribution of existing data and crime recognition. They proposed

a model in which the analysis and prediction of crimes are done

through the optimization of outlier detection operator parameters

which is performed through the Genetic Algorithm. The features

are weighted in this model and the low-value features were

deleted through selecting a suitable threshold. After which the

clusters are clustered by the k-means clustering algorithm for

classification of crime dataset.

Satyadevan et al. [5] has done a work which will display high

probability for crime occurrence and can visualize crime prone

areas. Instead of just focusing on the crime occurrences, they are

focusing mainly on the crime factors of each day. They used the

Naïve Bayes, Logistic Regression and SVM classifiers for

classification of crime patterns and crime factors of each day.

Their method consists of a pattern identification phase which can

identify the trends and patterns in crime using the Apriori

Algorithm. The prediction of crime spots is done with the help of

Decision Tree algorithm which will detect the crime possible

areas and their patterns.

Bruin et al. [7] proposed a technique which is used to

determine the clustering of criminals based on the criminal

careers. The criminal profile per offense per year is extracted from

the database and a profile distance is calculated. After that, the

distance matrix in profile per year is created. The distance matrix

including the frequency value is made to form clusters by using

naïve clustering algorithm. They made a criminal profile which is

established in a way of representing the crime profile of an

offender for a single year. With this information, the large group

of criminals is easily analyzed and they predicted the future

behavior of individual suspects. It will be useful for establishing

the clear picture on different existing types of criminal careers

They tested the tool on actual Dutch National Criminal Record

Database for extracting the factors for identifying the criminal

careers of a person.

3.3 SPATIAL AND GEO-LOCATION BASED METHODS

Huang et al. [6] focused on a different approach for criminal

activity prediction based on mining location based Social

Network interactions. By using these interactions, they can collect

information using the geographical interactions and data

collections from the people. They devised a working procedure in

which a series of features are categorized from the Foursquare and

Gowalla used in the San Francisco Bay area. The crime patterns

and the crime occurrences are tracked with the geographical

features which are extracted from the map and they are analyzed

to detect the urban areas with high crime activities. Their work

aims at exploiting the location-based social network data to

investigate the criminal activities in urban areas. By using the

Haversine formula the distance between the two points i.e. the

crime location and venue location is calculated and shown in the

Google Maps API and OpenStreetMap.

Chen [19] have presented a general framework for crime data

mining that draws on experience gained with the Coplink project

with the researchers at Arizona and their work mainly focuses on

showing the relationships between crime types and the link

between the criminal organizations. They used a concept space

approach which will extract criminal from the incident summaries.

Yu [20] have discussed the preliminary results of a crime

forecasting model developed in collaboration with the police

department of a United States city in the Northeast. Their approach

is to architect datasets from original crime records. The datasets

contain aggregated counts of crime and crime-related events

categorized by the police department. The location and time of

these events is embedded in the data. Additional spatial and

temporal features are harvested from the raw data set. Second, an

ensemble of data mining classification techniques is employed to

perform the crime forecasting. Then they analyzed a variety of

classification methods to determine which is best for predicting

crime “hotspots”. They even investigated classification on increase

or emergence. Last, they have proposed the best forecasting

approach which is aimed at achieving the most stable outcomes.

Rizwan et al. [22] have performed classification of crime

dataset to predict Crime Category for different states of the United

States of America. The crime dataset that they used in this research

is real in nature. That is, it was collected from socio-economic data

from 1990 US Census, law enforcement data from the 1990 US

LEMAS survey, and crime data from the 1995 FBI UCR. Their

work compared the two different classification algorithms namely,

Naïve Bayesian and Decision Tree for predicting Crime Category

for different states in USA. The results from their experiment

showed that, Decision Tree algorithm out performed Naïve

Bayesian algorithm and achieved 83.9519% Accuracy in

predicting Crime Category for different states of USA.

Donald [24] have proposed a system for Crime Analysis which

was named by them as The Regional Crime Analysis Program

(ReCAP) system. It was designed by them as a computer

application designed to aid local police forces (e.g. University of

Virginia (UVA), City of Charlottesville, and Albemarle County)

in the analysis and prevention of crime. ReCAP works in

cooperation with the Pistol 2000 records management system,

which aggregated and housed all of the crime information from a

region. Their research and development was primarily focused on

the individual components of the system which includes a

database, geographic information system (GIS), and data mining

tools which consisted of data mining alrogithms which produced

spatial mining results over the crime hotspots. Their system

consists of the seamless integration of all the components in the

system.

3.4 PRISONER BASED METHODS

Sheehy et al. [10] came up with a research idea which was

geared towards the treatment of the mentally ill people inside the

prison. According to their work, the mentally ill criminals are

identified using their Social Security Number (SSN) with all the

criminal personal records and their crime career records attached.

As the outcome, the Criminals are classified into “high”,

“medium” and “low” levels of recidivism risk potential according

H. BENJAMIN FREDRICK DAVID AND A. SURULIANDI: SURVEY ON CRIME ANALYSIS AND PREDICTION USING DATA MINING TECHNIQUES

1462

to their mental health. Their objective was to describe and classify

the criminals into a misdemeanor and a felony which can be

referred and not referred based on the mental health of the

criminals. Their ill activities are monitored and data collection is

continuous. By these, the criminals can be separated from other

criminals who are hazardous and those who can cause damage to

other inmates along with them. Further, their study also involves

the classification of the mental health of the criminals into two

categories i.e. “referred” and “not-referred”. This helps the guards

to identify the prisoners who are referred for the mental health

check-up. The research work they had undergone will provide a

summary of the inmates who are seriously mentally ill and those

who are to be separated from the other inmates.

3.5 COMMUNICATION BASED METHODS

Taha et al. [9] has developed a forensic investigation tool for

identifying the influential members who create an impact in a

criminal organization. The immediate leaders can also be

identified in a criminal organization. Removing these influential

members can weaken the strength of the criminal organization.

Their work is based on this methodology. They proposed a new

work which is known as SIIMCO which first constructs the graph

representing the criminal group or organization as a network from

either mobile communication data of the criminal organization or

based on the crime records. The system works on the basis of the

created networks. These networks represent the criminal

organization or crime incident reports. The vertex represents the

individual criminals and the link represents the relationships or

communication link between those two criminals. They employed

certain formulas that quantify the degree of influence/ importance

of each vertex in the network relative to all other vertices i.e.

criminals in the graph. Based on this their system identifies the

immediate leaders with the weighted graph which connects the

criminals and identify them for further processing.

4. QUALITATIVE ANALYSIS OF CRIME ANALYSIS AND PREDICTION

APPROACHES

The prediction can be made based on the Textual information

or the Geospatial information or even the prisoner records which

were manually recorded. By using the real open data such as

internet, social feeds and messages the researcher can use the text

processing or NLP techniques to mine information from the data

and categorize the e-mails, messages or posts into a suspicious or

a non-suspecious record [1]. Whereas in the Spatial mining area,

the extraction of features from SNAP Gowalla dataset, DataSF

criminal dataset up to February 2015 provides the way to plot the

crime occurences on the Google Map which is interpreted easily.

The communication based methods describe the identification of

the leaders in a criminal organization may be a tedious process.

Kamal Taha et al. [9] produced an approach through the phone

calls and other communication data such as call logs and records,

the influential members on a crime organization can be tracked.

Kevin Sheehy, Thomas Rehbreger, Andrew O’Shea, William

Hammond, Charlotte Blais, Michael Smith K., Preston White, Jr.,

Neal Goodloe [10] introduced an approach to categorize and

identify the mentally ill prisoners among the prisoners and keep

them separate from other prisoners to avoid conflict and injuries

between them. Even though there are many methods for analyzing

the crimes , this paper concludes many results based on the

qualitative analysis When considering the Text/NLP based

methods, Hamdy et al. [8] overcame the defects from the work of

Sharma [1] based on many factors such as implementation of

preprocessing for the data and extraction of relevant features.

Both the paper labels the outcome based on suspicious activity.

Mugdha Sharma [1] used an enhanced ID3 algorithm whereas the

work produced by Ehab Hamdy, Ammar adl, Aboul Ella

Hassanien, Osman Hegazy and Tai-Hoon Kim. [8] does not

specifies the classification algorithm. The weakned of this paper

is mostly about not giving the clear view of the pre processing and

classification alogorithm. When considering crime patterns and

evidence based methods, there are clustering and classification

based papers. Bogahawatte and Adikari [2] concentrated on using

the Naïve Bayes for finding out most possible suspect. Jyoti

Agarwal et al. [3] on the other hand focused on crime analysis by

implementing the K-Means clustering algorithm on crime dataset

using rapid miner tool and the author had performed the crime

analysis by considering the homicide crimes and plotting it with

respect to year. Kiani et al. [4] concentrated on using the Genetic

Algorithm to optimize the distance operator parameter of the

decision tree using GINI index. The clustering of the criminal

careers has been effectively done in the work [7]. Whereas, Shiju

Satyadevan, et al. [5] have performed a comparison of the Naïve

Bayes, SVM, Logistic Regresion and Decision Tree. This paper

presents the crime prone regions and represented as heatmaps

which indicate the level of heat. When considering the Spatial and

Geolocation based methods, all these methods are analysed based

on qualitative manner and the analysis information is described in

the below mentioned table Table.1.

Table.1. Qualitative Analysis of Crime Analysis and Prediction

METHOD INPUT DATASET

USED

PRE

PROCESSING

FEATURE

EXTRACTION

CLASSIFICATION/

CLUSTERING STRENGTH WEAKNESS OUTCOME

Text/

NLP-

based

methods

[1] E-mail

messages

Real and

open emails

sent by

terrorists

and some

are dummy

emails

Nil

Selection of a

subset of the

original text

containing “kill”,

“death”, “bomb”,

“guns”, “blasts”

Enhanced ID3

Decision Tree

algorithm

Introducing

attribute

importance as

a factor before

information

gain in the

decision tree

Nil

Labeling email

as Suspicious,

Non-

suspicious, and

May be

suspicious

[8]

Crime

history, age,

previous

Device

sensors,

Security

Structuring

collective data

into {Time,

Similarity

matching for

sensory images

A trained

classification model

is used to predict the

Consideration

of location

feeds and

Not giving a

clear view of

the processing

Suspicious

behavior to

three levels

ISSN: 2229-6959 (ONLINE) ICTACT JOURNAL ON SOFT COMPUTING, APRIL 2017, VOLUME: 07, ISSUE: 03

1463

arrests,

Modus

Operandi,

countries

visited,

place of

birth,

Average use

of ATM,

Types of

crimes,

Entrance

with respect

to Time of

Day, Crime

areas,

Victims’

mistakes

camera

information,

Messages,

Audio feeds,

Social

network

posts and

messages

Final

Movement,

Frequency rate,

Video, Images,

Audio }

using sliding

window. Text

semantic Analysis

of the text

information

performed using

Lexical processing,

Natural Language

Processing (NLP).

similarity of a given

input to the

suspicious item or

location.

mobile usage

information

and

comparison of

criminal

behavior.

such as

“High”,

“Medium” and

“Low”

Crime

patterns

and

Evidenc

e-based

methods

[2]

Crime

evidences

including

many

attributes

like crime

scene, day,

month,

offense,

resources

used, time,

role in

crime,

transportatio

n etc.,

Colombo

crime and

criminal

records

Nil Extraction of

evidence

Clustering based

model to identify

patterns of

committing crimes.

Naïve Bayes

classifier applied to

find most possible

suspect

Uses Naïve

Bayes so this

can be even

suitable for

small datasets.

No clear view

of clustering

method and

Prisoner

verification

Finding

Categories as

robbery,

burglary, and

theft

Classifying

person as

“suspect” and

after judgment

“criminal”

[3]

Homicide

crimes and

their

occurrences

Crime

dataset for

crime

analysis by

polices in

England and

Wales from

1990 –

2011-12

Nil

Extraction of crime

patterns based on

the available crime

and criminal data

K-means clustering

algorithm

Produces year

wise clusters

of homicide

crimes

committed

Concentration

is only on

clustering of

homicide

crimes

Year and

analysis of

variation in

clusters formed

[4]

Burglary,

Robbery,

and

Homicide

Crime

dataset for

crime

analysis by

polices in

England and

Wales from

1990 – 2011

Nil

Filtering of

dataset, Outlier

detection using

distance operator

(k-NN), Genetic

Algorithm used for

optimizing of

outlier detection

operator

parameters

Classification was

done using Decision

Tree using GINI

index and the testing

and training done

using Sample

Stratified

Use of GA to

optimize the

distance

operator

parameters in

Clustering and

Predict the

cluster’s

members

based on

classification

using Decision

Tree

The number of

clusters in the

clustering

process needs

to be

optimized and

further

optimization

of the

technique

needs to be

done

The results for

the optimized

and non-

optimized

parameters

were compared

to show the

difference in

quality and

effectiveness

[5]

location,

date, type of

crime data

extracted

from

Websites,

Blogs,

Social

Media, RSS

Feeds

Websites,

Spatial

Information,

and date

about crimes

Nil

Extraction of the

following crime

data related to

“vandalism”,

“murder”,

“robbery”,

“burglary”, “sex

abuse”, “gang

rape”, “arson”,

“armed robbery”,

Naïve Bayes, SVM,

Logistic regression

Crime prediction was

done using decision

tree which is done

using sample police

complaints

Comparison of

Naïve Bayes

with SVM.

Decision Tree

is easy to

interpret and

understand for

crime spot

identification.

Not predicting

the time in

which the

crime is

happening.

The crime-

prone areas

(regions) are

graphically

represented

using a heat

map which

indicates the

level of crimes

H. BENJAMIN FREDRICK DAVID AND A. SURULIANDI: SURVEY ON CRIME ANALYSIS AND PREDICTION USING DATA MINING TECHNIQUES

1464

“highway

robbery”,

“snatching”

[7]

Crime

database and

criminal

information

National

Crime

Record

Database

Nil

Crime nature,

frequency,

duration, severity

Crime profile of

offender for single

year is determined for

comparison and he

Development

of new

distance

measures with

combination

of profile

distance with

crime

frequency of

criminals

The runtime

of the chosen

approach is

not optimal

Clustering of

criminal

careers based

on the nature.

One time

criminal,

severe

criminals and

minor career

criminals

Spatial

and

Geo-

location

based

methods

[6]

Geo-

location and

Crime Type

SNAP

Gowalla

dataset,

DataSF

criminal

dataset up to

February

2015

Extraction of

crime type like

Assault,

Robbery, Theft,

Vandalism,

Drug

Geographical

features,

Popularity,

Location category,

Neighbor entropy,

Social Tightness

density, crime

location, venue

from Foursquare

Random Forest(RF),

Linear Regression

(LR) and Support

Vector Machine

(SVM)

Random Split

method

utilized with

80% for

training and

20% for

testing in

classification

Nil

Crime Areas

plotted using

Google Map

API and

OpenStreetMa

p in San

Francisco Bay

area and

Criminal

pattern

discovery

according to

the context of

user activity

and location-

based social

networks.

Predict crime

frequency and

find which

crime is to be

more difficult

or easier to be

predicted

Commu

nication

based

methods

[9]

Flow of

communicat

ions/informa

tion links

between two

criminals

(e.g., phone

call records,

messages,

etc.), names

of

criminals/su

spects, the

type of

crime,

location and

date of the

crime.

Real-world

communicat

ion records

(DBLP,

Enron email

dataset,

Nodobo

mobile

phone

records

dataset)

Creating the

graph based on

the data and

then assigning

weight to a

vertex based on

its number of

communication

attempts in the

criminal graph

The immediate

leaders of lower-

level criminals and

the lower-level

criminals

themselves are

extracted.

Evaluation of the

accuracy of the three

systems by measuring

their Recall,

Precision, and

Euclidean Distance.

Evaluated

SIIMCO by

comparing it

experimentall

y with

CrimeNet

Explorer and

LogAnalysis

Nil

System can

identify the

influential

members of a

criminal

organization

and the

immediate

leaders of a

given list of

lower-level

criminals

Prisoner

based

methods

[10

]

The Social

Security

Number

(SSN) with

all the

criminal

personal and

crime career

records.

Albemarle-

Charlottesvil

le Regional

Jail (ACRJ),

Jefferson

Area

Community

Corrections

(JACC) and

A combination

which includes

the Social

Security

Number (SSN)

and date was

used to link the

databases

together.

age, criminal

history,

employment

history, crime

type := “assault”,

“larceny”,

“supervision

violations”,

“narcotics

Offenders are

classified into three

classes namely

“high”, “medium”,

and “low” as levels of

recidivism risk

potential. Further, the

mental health status

of the inmates is

Analysis for

the

identification

of the

mentally ill

felony.

Statistical

classification

of criminals

missing.

Could have

taken more

features

“Referred”

individuals can

be made to

have a longer

stay in jail

longer than

“not-referred”

individuals.

ISSN: 2229-6959 (ONLINE) ICTACT JOURNAL ON SOFT COMPUTING, APRIL 2017, VOLUME: 07, ISSUE: 03

1465

Region Ten

Community

Services

Board.

charges”, “traffic

violations”,

“driving while

intoxicated”,

categorized into two

categories “referred.”

and “not-referred.”

5. QUANTITATIVE ANALYSIS OF CRIME ANALYSIS AND PREDICTION

APPROACHES

For performing the quantitative analysis of the methods taken,

the performance metric value needed to be computed and they are

to be compared with the other. Hence, for performing the

calculations of the performance metric there are a few formulas

which can be utilized for achieving the performance value from

the dataset. The formulae for the calculation of the performance

metrics are given below in Table.2.

Table.2. Metrics and their formula

METRIC FORMULA

Accuracy FNFPTNTP

TNTP





Error Rate 100 - Accuracy

Precision FPTP

TP



Recall FNTP

TP



F-value  

  

 

RP

11

2

where, P is the Precision and R is the Recall

Although many papers were studied in the literature review all

the papers were irrevalent to the crime prediction and criminal

analysis domain. Hence a few papers in the crime analysis and

prediction domain has been taken and their results have been

reproduced as given originally in the reference papers. The below

given table Table 3 provides quantitative analysis of the three tools

and the Decision Tre algorithm which is suppored with the Genetic

Algorithm for optimization of the parameters. When the parameters

are optimized, the classification accuracy of the Decision Tree is

increased a bit further. This shows that although the Decision tree

performs well, when it used with the Genetic Algorithm for

optimization of the decision tree paramters, the results shown show

significant improvement in the accuracy and further more the tools

given below have the metric value, which is purely based on the

dataset and the records and the performance values are taken as it

is in the reference paper. The quantitative analysis produced results

which show the increase in the accuracy level of classification

because of using the GA to optimize the parameters. This occurs

because of the ability of the GA to learn the optimal values and then

it is applied to set the parameter to optimal value when performing

calculation. Also, the Precision, Recall and F-value varies from the

dataset and the system. This shows the SIIMCO performing well

when defined in terms of the metrics.

Table.3. Quantitative Analysis of Crime Analysis & Prediction

NAME OF

THE

METHOD/

SYSTEM

PERFORMANCE

METRIC

PERFORMANCE

VALUE

[4]

Decision

Tree

classification

with GA for

optimizing

the the

parameters

Accuracy of

Prediction

Optimized

parameter 91.64%

Non-

Optimized

parameter

85.74%

Clasification Error

Optimized

parameter 8.36%

Non-

Optimized

parameter

13.26%

Fitness Function

Optimized

parameter 72.28%

Non-

Optimized

parameter

72.48%

[9]

SIIMCO

Recall 0.62

Precision 0.56

F-Value 0.59

CrimeNet

Explorer

Recall 0.36

Precision 0.41

F-value 0.38

Log Analysis

Recall 0.53

Precision 0.51

F-value 0.52

6. CONCLUSION

In this paper, we have studied some known approaches for

crime analysis and prediction concerned with data mining.

Although many papers have been studied, only those papers with

background in the crme prediction and criminal identification

papers are compared with a theoretical study. Each paper has their

own advantages and disadvantages. Each paper has its own

individual approach for solving the crimes and criminal

prediction. This is a theoretical study for several methods in

identification of crime and criminals which includes Text/ NLP

based methods, crime patterns and crime evidence based methods,

spatial and geo location based methods, communication based

methods and finally Prisoner based methods. The data mining

techniques studied from this survey can be applied for identifying

the criminals in the society and also for providing a better future

to live in.

H. BENJAMIN FREDRICK DAVID AND A. SURULIANDI: SURVEY ON CRIME ANALYSIS AND PREDICTION USING DATA MINING TECHNIQUES

1466

REFERENCES

[1] Mugdha Sharma, “Z-Crime: A Data Mining Tool for the Detection of Suspicious Criminal Activities based on the

Decision Tree”, International Conference on Data Mining

and Intelligent Computing, pp. 1-6, 2014.

[2] Kaumalee Bogahawatte and Shalinda Adikari, “Intelligent Criminal Identification System”, Proceedings of 8th IEEE

International Conference on Computer Science and

Education, pp. 633-638, 2013.

[3] Jyoti Agarwal, Renuka Nagpal and Rajni Sehgal, “Crime Analysis using K-Means Clustering”, International Journal

of Computer Applications, Vol. 83, No. 4, pp. 1-4, 2013.

[4] Rasoul Kiani, Siamak Mahdavi and Amin Keshavarzi, “Analysis and Prediction of Crimes by Clustering and

Classification”, International Journal of Advanced

Research in Artificial Intelligence, Vol. 4, No. 8, pp. 11-17,

2015.

[5] Shiju Sathyadevan, M.S. Devan and S. Surya Gangadharan, “Crime Analysis and Prediction using Data Mining”,

Proceedings of IEEE 1st International Conference on

Networks and Soft Computing, pp. 406-412, 2014.

[6] Yu-Yueh Huang, Cheng-Te Li and Shyh-Kang Jeng, “Mining Location-based Social Networks for Criminal

Activity Prediction”, Proceedings of 24th IEEE International

Conference on Wireless and Optical Communication, pp.

185-190, 2015.

[7] Jeroen S. De Bruin, Tim K. Cocx, Walter A. Kosters, Jeroen F. J. Laros and Joost N. Kok, “Data Mining Approaches to

Criminal Career Analysis”, Proceedings of 6th IEEE

International Conference on Data Mining, pp. 1-7, 2006.

[8] Ehab Hamdy, Ammar Adl, Aboul Ella Hassanien, Osman Hegazy and Tai-Hoon Kim, “Criminal Act Detection and

Identification Model”, Proceedings of 7th International

Conference on Advanced Communication and Networking,

pp. 79-83, 2015.

[9] Kamal Taha and Paul D. Yoo, “SIIMCO: A Forensic Investigation Tool for Identifying the Influential Members

of a Criminal Organization”, IEEE Transactions on

Information Forensics and Security, Vol. 11, No. 4, pp. 811-

822, 2016.

[10] Kevin Sheehy et al., “Evidence-based Analysis of Mentally 111 Individuals in the Criminal Justice System”,

Proceedings of IEEE Systems and Information Engineering

Design Symposium, pp. 250-254, 2016.

[11] David J. Hand, Heikki Mannila and Padhraic Smyth, “Principles of Data Mining”, MIT Press, 2001.

[12] 10 Emerging Technologies That Will Change Your World, Available at: http://www.rle.mit.edu/thz/documents/10_emerging_tech.p

df.

[13] U. Fayyad, G. Piatetsky-Shapiro and P. Smyth, “The KDD Process for Extracting Useful Knowledge from Volumes of

Data”, Communications of the ACM, Vol. 39, No. 11, pp.

27-34, 1996.

[14] Illhoi Yoo, Patricia Alafaireet, Miroslav Marinov, Keila Pena-Hernandez, Rajitha Gopidi, Jia-Fu Chang and Lei Hua,

“Data Mining in Healthcare and Biomedicine: A Survey of

the Literature”, Journal of Medical Systems, Vol. 36, No. 4,

pp. 2431-2448, 2011.

[15] Shyam Varan Nath, “Crime Pattern Detection using Data Mining”, Proceedings of IEEE/WIC/ACM International

Conference on Web Intelligence and Intelligent Agent

Technology Workshops, pp. 1-4, 2006.

[16] Hsinchun Chen, Wingyan Chung, Yi Qin, Michael Chau, Jennifer Jie Xu, Gang Wang, Rong Zheng and Homa

Atabakhsh, “Crime Data Mining: An Overview and Case

Studies”, Proceedings National Conference on Digital

Government Research, pp. 1-5, 2003.

[17] Tong Wang et al., “Learning to Detect Patterns of Crime”, Joint European Conference on Machine Learning and

Knowledge Discovery in Databases, pp. 515-530, 2013.

[18] Karl F. Schuessler and Donald R. Cressey, “Personality Characteristics of Criminals”, American Journal of

Sociology, Vol. 55, No. 5, pp. 476-484, 1950.

[19] H. Chen, W. Chung, J.J. Xu, G. Wang, Y. Qin and M. Chau, “Crime Data Mining: a General Framework and Some

Examples”, Computer, Vol. 37, No. 4, pp. 50-56, 2004.

[20] Chung-Hsien Yu, Max W. Ward, Melissa Morabito and Wei Ding, “Crime Forecasting using Data Mining Techniques”,

Proceedings of 11th IEEE International Conference on Data

Mining Workshops, pp. 779-786, 2011.

[21] P. Thongtae and S. Srisuk, “An Analysis of Data Mining Applications in Crime Domain”, Proceedings of IEEE 8th

International Conference on Computer and Information

Technology Workshops, pp. 122-126, 2008.

[22] Rizwan Iqbal, Masrah Azrifah Azmi Murad, Aida Mustapha, Payam Hassany Shariat Panahy and Nasim

Khanahmadliravi, “An Experimental Study of Classification

Algorithms for Crime Prediction”, Indian Journal of Science

and Technology, Vol. 6, No. 3, pp. 4219-4225, 2013.

[23] Shyam Varan Nath, “Crime Data Mining”, Proceedings of Advances and Innovations in Systems, Computing Sciences

and Software Engineering, pp. 405-409, 2007.

[24] Donald E.Brown, “The Regional Crime Analysis Program (RECAP): A Framework for Mining Data to Catch

Criminals”, Proceedings of IEEE International Conference

on Systems, Man, and Cybernetics, pp. 2848-2853, 1998.

[25] Colleen McCue, “Data Mining and Predictive Analysis: Intelligence Gathering and Crime Analysis”, Butterworth-

Heinemann, 2014.

[26] Arunima S. Kumar and Raju K. Gopal, “Data Mining based Crime Investigation Systems: Taxonomy and Relevance”,

Proceedings of Global Conference on IEEE Communication

Technologies, pp. 850-853, 2015.

[27] Manish Gupta, B. Chandra and M.P. Gupta, “Crime Data Mining for Indian Police Information System”, Journal of

Crime, Vol. 2, No. 6, pp. 43-54, 2006.