SARIMA-GRU Crime Prediction Model Based on Nonlinear Combination of BP Neural Network

Abstract

Aiming at the problems that the current crime prediction model can not capture the composite characteristics of crime sequence data or respond to the dynamic changes of the environment in time, a SARIMA-GRU crime prediction model based on nonlinear combination of BP neural network is proposed. This model nonlinearly combines the prediction results of SARIMA and GRU models on the number of crimes through BP neural network, uses the back-propagation algorithm to learn the weight, and takes the weight matrix determined by each layer of neurons as the weight of the two methods in the combined model. Comprehensive utilization is taken of the advantages of SARIMA model in linear time series prediction and GRU in nonlinear feature mining, so as to obtain better crime prediction results. Through the open crime data of Vancouver and San Francisco to compare the combined model with other models, the experimental results show that the combined prediction model proposed in this paper can capture the composite characteristics of crime time series data, and has higher accuracy than other crime prediction models.

Keywords

• crime prediction model
• bp neural network
• gru
• sarima model
• combination forecast