A method of crime rate forecast based on wavelet transform and neural network

Abstract

Accurate prediction of crime is highly challenging. In order to improve efficiency of situational crime prevention, the temporal distribution of the crime rate within 24 hours was analysed and a forecast model combining discrete wavelet transform and resilient backpropagation neural network (DWT-RBPNN) is presented. First, historical crime incidence sequences obtained by the sliding window were decomposed by discrete wavelet transform. Then RBPNN trained decomposition sequences to predict the incidence of future trends and details. Finally, the trends and details were reconstructed to get the final prediction sequence. The experimental results showed that the proposed model has relatively high accuracy and feasibility on the crime rate prediction compared with single method of BPNN. The utility of the DWTRBPNN model can offer an exciting new horizon to provide crime rate forecasting and early warning in the situational crime prevention.