Prediction of Crime in Neighbourhoods of New York City using Spatial Data Analysis

Abstract

Crimes prediction is one of the most important topics in recent years that aim to protect people’s lives. These analytical studies for criminal hotspots are frequently demanded by law enforcement agencies hence, there is a huge requirement and demand for enhanced geographic information systems and innovative spatial data mining techniques in order to enhance crime detections and better protect their communities. In this paper, we propose a methodology to predict Spatio-temporal criminal patterns within the New York City neighbourhoods using a dataset from 2006 until 2019 with 2.2M criminal records for 25 different crimes type. In order to achieve the study objectives, the methodology passes through several stages until the final results are reached, starting with the visualization analysis of Spatio-temporal New York crime data which is important in decision-making, followed by, applying three different classifiers namely; Support Vector Machine (SVM), Random Forest (RF), and XGboost classifiers. After analysis, it is illustrated that XGboost has predicted the highest number of correct classifications out of 25 different crime types it has predicted 22 types of crime accurately, whereas Random Forest has predicted 21 types of crime accurately and SVM predicted accurately 17 types of crimes with lowest accuracy. Hence XGBoost outperformed all other models and can be considered for detection of crimes in the neighborhood.