Comparative Study of Machine Learning Models for
Crime Analysis
Yashvi Mehta
1
, Sakshi Mahadik
2
, Manan Shah
3
, Jaiwin Shah
4
and Dr. Sudhir Dhage
5
1-5
Computer Engineering Sardar Patel Institute of Technology, Mumbai, India
Email: yashvi.mehta@spit.ac.in, sakshi.mahadik@spit.ac.in, manan.shah@spit.ac.in, jaiwin.shah@spit.ac.in,
sudhirdhage@spit.ac.in
Abstract—In today’s world where criminal activity is increas- ing daily, it’s necessary to curb
these crimes. Analysis of crime is a procedural approach to identifying crimes. Crimes can
reduce if the criminal hotspots get identified. If these hotspots are pinned down based on the
crime type, crime detection and analysis can. With the machine learning techniques applicable
to big datasets, crime investigators could use these approaches to narrow down their search
and solve the cases. We intend to preprocess the data and find these criminal hotspots to
improve our search. In this paper, we have implemented several machine learning algorithms
such as decision tree regression, linear regression, random forest, and other algorithms to
analyze crime patternsin the United States. A comparative study is done to see which algorithm
would yield better results in terms of accuracy.
Index Terms— Crime Analysis,Decision Tree Regression, K- Nearest Neighbors , Linear
Regression, Support Vector Machine, Random Forest.
I. INTRODUCTION
Criminal activities are gradually rising around us. To reduce them, high crime rate areas need to be determined.
The first step is to segregate different categories of crimes committedin that neighborhood. The main purpose
of crime analysis is to reduce these organized criminal activities. Crime prediction can be employed by
identifying these hotspots and reporting them to the concerned authority, thereby helping to decrease the crime
rate. It can be done by taking the help of various machine learning models. It would utilize the existing crime
data and predict the crime type and its occurrence. The major goal behind this research is to identify crimes
that can get predicted once the required data gets filtered out. This filtered data would lead to the finding of
patterns that would come in handy to identify the criminal activities in a state.
The dataset taken consists of a variety of attributes. After gathering the data, the data is cleaned and divided into
train and test sets. Supervised techniques like K-nearest neigh- bour, Linear Regression, Decision Tree, Random
Forest, and Support Vector Machine are used for crime prediction. The results from this prediction can be very
useful for the police department in investigating cases and reducing the crime rate by taking appropriate
measures which will secure the people and the region.
II. LITERATURE SURVEY
Akash Kumar, Aniket Verma, Gandhali Shinde, Yash Sukhdeve, and Nidhi Lal [1] proposed a crime prediction
Grenze ID: 01.GIJET.9.1.31
© Grenze Scientific Society, 2023
Grenze International Journal of Engineering and Technology, Jan Issue