Analysis And Application Of Crime Hot Spot Data Based On Temporal And Spatial Features

The advent of the information age has affected and changed the paradigm of policing,new requirements for crime prevention intervention have been introduced,and the hot area of predictive policing is attributed to crime hotspot prediction.Based on the real data set of City H,it is hoped to improve the prediction effect of crime hotspots.The main work is as follows:(1)To develop analysis and forecast feasibility judgment from the perspective of data analysis.Firstly,the data set is preprocessed and selected,and then analyzed separately from the perspective of space and time,and further judged for the feasibility of prediction.(2)From a spatial perspective,explore the impact of the addition of covariates on the prediction of crime hotspots in City H.Firstly,the map is mapped according to the hotspot distribution,and the map is divided into four categories by using a grid and dividing the grid information into four categories by clustering,and then the selected covariates are added to the model for experiments.The results show that the inclusion of covariates has a positive effect on the prediction and the different choices of covariates have different effects on the prediction.(3)Develop forecasts from temporal characteristics.Autoregressive Integrated Moving Average model(ARIMA)is commonly used in time series forecasting,but it is more suitable for dealing with linear data,Long Short Term Memory Network(LSTM)has a strong advantage in dealing with nonlinear data,and the combined ARIMA-LSTM model can be constructed to fully exploit the data information to improve the forecasting accuracy.Firstly,the ARIMA model is used to obtain the prediction results of linear data in the data and the filtered residual series which is nonlinear data,then the LSTM model is used to obtain the prediction results of nonlinear data,and finally the linear prediction results and nonlinear prediction results are explored with the LSTM model to get the final prediction results.The results demonstrate that the combined ARIMA-LSTM model is more effective than the single ARIMA model and the LSTM model in predicting property crime in X district of H city,and the combined model can better fit the actual case trends.(4)Adding spatial features to develop predictions from the perspective of spatiotemporal features.Inspired by traffic flow prediction,the Spatio-Temporal Graph Convolutional Networks(STGCN)model is used to capture the spatio-temporal correlation on crime prediction.The data set of X district of H city is processed.Spatially,the area is divided by streets to construct the graph structure information,the points of the graph represent the streets,the edges represent the adjacency between streets and construct the spatial matrix;temporally,the property crime data of each street in X district is divided by days,and finally the spatio-temporal data is constructed into threedimensional data for storage.The spatio-temporal feature data is processed by a spatiotemporal convolution block,and a prediction result based on the spatio-temporal feature is obtained.The results show that for the same dataset,the prediction effect of adding spatio-temporal features is to some extent higher than that of considering only temporal features,and the STGCN model works well in predicting property-based crimes in X district of H city,and the model is also applicable to predicting violent crimes in X district of H city.