Design And Implementation Of Crime Early Warning System Based On Machine Learning Algorithm

The global economy,politics and culture are developing rapidly.Crime is an obstacle to these developments and endangers people's property safety.However,traditional prediction methods cannot accurately predict crime.The paper predicts crime based on machine learning algorithms.Crime prediction refers to the prediction of past crime records,so as to predict the hot spots,quantities,types,etc.of crimes in the next period of time.Therefore,the following research has been carried out on the number of regional crimes,crime hotspots and crime types.The paper introduces the recurrent neural network to predict the number of regional crimes,and establishes a regional crime prediction model based on LSTM algorithm.Two implementation methods are proposed:LSTM one-dimensional time series and LSTM multi-dimensional time series.The study area is meshed to form a small grid area,and then the target grid is selected.The traditional time series algorithm ARIMA only uses the historical crime data of the target grid,and LSTM also utilizes the historical crime of the grid around the target grid.The data considers the impact of the number of surrounding grid crimes on the number of target grid crimes.The paper is based on the historical crime dataset of Chicago.The results show that the RMSE of LSTM is 0.73 lower than that of ARIMA,and the MAPE of LSTM is 4.96%lower than that of ARIMA.Furthermore,the paper improved the LSTM and incorporated the holidays,temperature and weather factors into the model.The results show that the RMSE of the improved model is 0.57 lower than that of the unmodified LSTM,and the MAPE is 2.62%lower than that of the unmodified LSTM.Aiming at the regional hotspot prediction,this paper proposes a spatiotemporal neural network that considers the time and space of the target region embedded in the model.The model introduces the concept of spatio-temporal window,which will predict whether a region is a hot spot,and transform it into a spatio-temporal sequence prediction problem,based on the cumulative effect of time and space to predict whether a region is a hot spot.For the purpose of verifying the algorithm's validity,six classical classification algorithms such as decision tree,random forest and logistic regression,are used to predict.Experimental results show that random forests have the best effect in traditional algorithms.The accuracy of the space-time model proposed by the paper is 5.5%higher than that of random forest,the accuracy rate is 6.5%higher than that of random forest,the recall rate is 6.9%higher than that of random forest,and the F1-score is 0.061 higher than that of random forest.The paper studies the data preprocessing process of crime type prediction.The excessive number of independent coordinate pairs in the original data set can not establish the model.The paper uses the clustering algorithm to map the huge independent latitude and longitude coordinates to 20 different regions,and the original is more The crime types are combined into three types,and then the classifier modeling in the sklearn library is called,and finally the probability of occurrence of each of the three crime types can be output according to the input location and time.