Time-Series Models for Crime Prediction in India

Abstract

The dark web is termed as the concealed content hidden behind interfaces offering secure chats or mailing applications on the WWW, and exploiting it for illegal practices is termed cybercrimes. As the dark web is enormously big compared to the surface web, cyber-crimes have become one of the biggest challenges to tackle. Predicting them before they actually happen is almost impossible. The work thus deals with predicting the number of cognizable crimes (physical and cybercrimes reported collectively) by National Crime Record Bureau (NCRB), the government-approved agency for maintaining crime records in India. Predicting crimes has become important as they advance to new levels of intelligence and diversification with the increased use of information and communication technology (ICT). This work presents the research based on evidences with ARIMA and LSTM time series models for predicting the cognizable crimes in India. The authors have used root mean squared error and mean absolute error as loss functions for error evaluation, and the results obtained from both the models are presented.