Spatio-Temporal Crime Analysis Using KDE and ARIMA Models in the Indian Context

Abstract

In developing countries like India, crime plays a detrimental role in economic growth and prosperity. With the increase in delinquencies, law enforcement needs to deploy limited resources optimally to protect citizens. Data mining and predictive analytics provide the best options for the same. This paper examines the news feed data collected from various sources regarding crime in India and Bangalore city. The crimes are then classified on the geographic density and the crime patterns such as time of day to identify and visualize the distribution of national and regional crime such as theft, murder, alcoholism, assault, etc. In total, 68 types of crime-related dictionary keywords are classified into six classes based on the news feed data collected for one year. Kernel density estimation method is used to identify the hotspots of crime. With the help of the ARIMA model, time series prediction is performed on the data. The diversity of crime patterns is visualized in a customizable way with the help of a data mining platform.