The Spatial-temporal Distribution Characteristics And Crime Prediction Of Three Classes Car Stolen

Crime not only threat to people's life and property safety,but also bring hidden dangers to social stability.Researchers have conducted a lot of researches in the field of criminology,so as to provide support and practical guidance for the work of police.Criminal geography is a branch of criminology which combines the theory of criminology with the knowledge of geography.Through analysis of the spatial-temporal distribution of crime,can discover criminal distribution characteristic and even predict crime trends.With the advence of ‘Big Data Time',researchers are not merely content with the knowledge presented in the chart,but pay more and more attention to explore the potential value of data.This will also bring changes to the work of police.This study takes car stolen cases of three classes including bicycle,electronic bicycle and moped in Changning district in Shanghai as the research object.Try to research the spatial-temporal distribution with the method of spatial aggregation,kernel density,association rules and then predict with support vector machine algorithm.Firstly,set different time scale(including month season,and 24 hours a day)and spatial scale(including community and roads)to do research on spatial distribution with standard deviation ellipse,kernel density and spatial aggregation.Then build the transaction table including crime road and crime period,discover knowledge with association rules.At last,utilize the support vector machine(SVM)to forecast the crime of each community during a period of time in the future.The study find that:(1)on the hand of temporal distribution,the amount of studied cases on seasonal scale decrease in winter,and then increase in spring,summer and fall season.The character of crime amount is more obvious in the month scale than the seasonal scale.It reflectsthat the quantity of the crime is of fluctuations in a season.The amount of crime within 24 hours have obvious periodicity.There were two peaks and troughs in a day,this is closely related to human activities.(2)on the hand of spatial distribution,most of the studied crimes concentrated in the northeast of research region.In addition,different season has different features of spatial agglomeration.Through the analysis of the amount of per length of the road can obviously find that the northern and southern roads is more than that of the eastern and western roads.This may be related to factors which can help escaping after crime.(3)Through the visualization of spatial-temporal association rules extracted find northwestern region has a higher amount of crime in daytime.The central region is easier to happen crime at night in spring,autumn and winter.There are a few roads happen crime during the day time and mainly concentrated in the summer.The eastern region is the prominent area in the spring and autumn season.During the two seasons the eastern region has become the prominent hour in the morning.And the crime mainly happened at night in summer mainly,while winter mainly during the day time.(4)SVM is a kind of data mining technology based on statistical theory whose core is the kernel function.In terms of nonlinear separable problems,take use of kernel functions to map the data set to high dimension space with principle of structure risk minimizing classification.The selected factors in this study including the total population of each community,the totally external population of each community,the distance to the nearest police station from each community,the number of crime in each community one month before,one month before whether the community had happened crime,the density of police patrol in each community.Select different factors to form a training data set and different kernel functions to construct the predictive model.By constructing the confusion matrix,use precision and recall rate to make comments on the generated model.Test themodel in the actual application scenario.The conclusion is gauss kernel function,an overall situation kernel function is stronger in generalization performance while is weaker in learning ability.And the polynomial kernel function,a local kernel function,is on the contrary.Relatively speaking,use the total population of each community,the totally external population of each community,the distance to the nearest police station from each community,the number of crime in each community one month before to construct a data set will get a relatively good effect in forecasting.