Study On Risk Analysis Of Urban Larceny Based On Multi-source Data

Larceny,as one kind of property crime,considerably threatens societal security,of which the crime rate is much higher than that of any other crime in China.Therefore,it is important to analyze the risk of larceny.Due to limited police resources,crime risk analysis based on data mining has gradually become an important method for risk prevention and control of the governments.However,the current data-based crime risk analysis methods often ignore the impacts of crime risk factors,thus leading to limited performance of prediction of future crime risk.Moreover,data sources for crime risk analysis are generally inadequate.Apart from that,the research schemes in different spatial scales have little difference and lack pertinence.In response to the above problems,this study uses the theft case data,the activity spots data of repeat offenders and community environment data in a large city in China to analyze the urban larceny risk by using multiple regression models,multiple classification models,and risk analysis methods.The contents are as follows:(1)Using multiple linear regression,machine learning nonlinear regression and geographically weighted spatial regression models,this study investigates the relation between activity spots of repeat offenders(and the CCTV distributions)and crime risk of larceny,and the results show that there is a clear correlation between them.The highest R~2of the three models of multiple linear regression,machine learning and geographical weighted regression are all higher than 0.59.And the three most important variables affecting the spatial distribution of larceny risk are“hotels and places of recreation where repeat offenders work”“the cybercafes which repeat offenders visit”and“CCTV distributions”.When it comes to the relationship between repeat offenders'activity spots and crime risk of larceny,the geographically weighted regression model has the best performance compared with the multiple linear regression model and the machine learning regression model(R~2=0.62).(2)Using multiple machine learning classification models,this study explores the prediction of the risk of crime consequences of burglary,pickpocketing and stealing electric vehicles based on the property of the case.The results show that classification models for predicting the consequence level of burglary,pickpocketing and theft of electric vehicle crime in the urban area has a good effect.And the F1-macro value predicted by the decision tree and the random forest model are all above 0.64 for each of the three types of cases,the F1-macro value of the consequences prediction of the theft of electric vehicles is as high as 0.7.In the three types of cases,“the area where the crime occurred”is the first in importance ranking,which is the most important variable affecting the predicting consequences of urban larceny.(3)Fuzzy analytic hierarchy process and fuzzy comprehensive evaluation method are used to assess the risk of community burglary.The community burglary risk assessment studies show that the construction and operation of“community police office”has the greatest impact on the risk of burglary,followed by“the security of key facilities”and“the status of residential space and home”.In the case study,the assessment results of the community burglary risk assessment model are basically consistent with the real situation of burglary,and the assessment effect is good.(4)The risk analysis software of larceny crime is designed and developed,which realizes the function of consequence prediction of urban larceny and risk assessment of community burglary.The models constructed in this study can better analyze the risk factors that affect the occurrence of larceny cases,predict the harmful consequences of theft cases and evaluate the risk of community burglary.It provides decision support and information tool support for intelligence analysis and risk prevention and control of larceny crimes.