Detecting and Predicting Crimes using Data Mining Techniques: Comparative Study

Crime is a major problem in our society where the
highest priority is concerned with individuals, society, and
government. Thus, it seems important to study factors and
relations between the occurrence of different crimes to avoid more
upcoming crimes. Crime prediction is a method of trying to study
the causes and motives of crime and predict the times and places
of its occurrence to reduce the commission of crimes that are
expected to occur in the future. Data mining is an important way
to facilitate the solution of future crime problems by investigating
hidden crime patterns and historical crime data. Therefore, this
study aims to analyze and discuss the various factors affecting the
commission of crimes and the methods that are applied to predict
future crimes and analyze their results. In this study, the model of
crime prediction is proposed which is based on some classification
algorithms such as (NB, KNN, Decision Tree, random forest,
Linear Regression, Logistic Regression, SVM), these classification
algorithms are applied to four real data sets (Chicago dataset, Los
Angeles dataset, Egypt dataset, United States dataset), Egypt data
set was extracted primarily from the online website
(Zabatak.com) and comparing between their scores. The
experimental results showed that the Random Forest classifier
achieves a high score on four data sets compared with other
classifiers. Random Forest achieves %88 on the Los Angeles
dataset, %92 on the Egypt dataset, %97 on the Chicago dataset,
and 81.7% on the United States dataset.