Exploring the Potential of AI for Urban Crime Prediction in India: A Case Study of Indian Metropolitan Cities

Urban crime prediction is of major interest to police departments and criminologists. This study focuses on the prediction of urban crime for 4 Indian cities (Surat, Delhi, Bangalore, Kolkata). Prediction is made for the total IPC crime rate for selected cities using the time period of 2001 to 2020. Here we have used a novel approach for prediction which uses the neural network time series application of MATLAB. This application has a nonlinear autoregressive tool that solves non-linear time series problems. So, this method can be referred to as NARNN (Nonlinear Autoregressive Neural Network). This method is compared with two classical methods of time-series forecasting (Linear regression, ARIMA). The results of the NARNN gave the highest accuracy for all 4 cities. It gave an accuracy of up to 98% for crime prediction in Delhi. When the crime rate is non-linear or doesn't have any significant trend ANN is strongly recommended. ARIMA is reasonably good when data have trends, while the use of linear regression for prediction is completely impractical.