LITERATURE REVIEW

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 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.

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