Study On Spatio-temporal Correlation Analysis And Prediction Methods For Urban Crime Cases

With the continuous increase in urban population and the accelerating process of urbanization,the number and types of urban crime cases are also increasing,which has brought tremendous pressure on urban security.Urban crime cases seriously affect the life and property safety of citizens,disrupt social order,and affect the social stability and economic development of cities.Therefore,crime spatial-temporal correlation analysis is of great significance and value for urban security management and crime prevention.This article analyzes the crime data from Chicago,USA,from 2012 to 2017,and applies relevant algorithm techniques to analyze and predict valuable scales such as the occurrence time,space,spatial-temporal correlation,and case categories.The main research results of the paper are as follows:(1)Analysis of the spatial-temporal distribution patterns of cases.The analysis of case time statistically analyzes the distribution of cases during daylight hours,and uses time series decomposition to analyze the changing trends of different types of cases on an annual and monthly scale.The analysis of case space uses the nearest neighbor index method to analyze the clustering of different crime geographical locations,and uses the KNN algorithm to generate a heat map of crime geographical locations,intuitively showing the clustering of crimes.(2)Analysis of the correlation between the time and space of cases.The FP-growth association rule mining algorithm is selected to perform crime spatial-temporal correlation analysis on three categories of case data.Different time periods are divided into day and night,and the locations are divided into grids of equal size.These attributes are treated as items,and the FP-growth algorithm is optimized to improve its efficiency in processing crime case data in time and space.Then,the optimized FP-growth algorithm is used to mine frequent itemsets,thereby obtaining the association rules between time and location,and identifying the spatial-temporal distribution patterns of crime.(3)Prediction of crime case categories.The Lightgbm,random forest,Adaboost,and Extra Trees algorithms are selected for crime prediction,and the predictive performance indicators of these four algorithms are analyzed.Then,the ensemble learning algorithm Stacking is used to combine these four algorithms through exhaustive permutation and combination.Through evaluation,the combination model with better performance than individual classification algorithms is identified,thus achieving better predictive performance in the task of predicting crime case categories.