Research On Data Mining Methods Of Spatiotemporal Hot Spots And Risk Factors Of Typical Violent Crimes

Violent crime refers to the aggressive crime behavior which adopts violence or coercion to achieve the criminals' purpose.Number of the violent crime incidents may directly reflect the social stability and level of public security management in a certain area.It has become a key step for effective prevention and response to violent crime that strengthens the monitoring of spatial-temporal hot spots,fully dig out the risk factors with the characteristics of cases and scientifically predict the consequences of the cases.Aiming at the problems of violent crime of single analysis method of spatial-temporal hot spots,strong subjectivity of risk assessment method,and lack of consequence analysis,considering randomness and serious consequences of violent crime and based on spatial-temporal hot spot analysis,word cloud analysis,and machine learning,this thesis proposes a new method coupled spatial-temporal hot spot analysis with risk analysis to meet the practical need of Ministry of Public Security on violence prevention.Using assault,robbery and rape data from city B,the proposed method is validated.The results indicate that:By spatial-temporal hotpots analysis,it is found that in terms of time,the violent crimes in city B fluctuated around the Spring Festival(declining before the Spring Festival and rising after Spring Festival),and the temporal hotspot was between 9.00 pm and 12.00 am.In total,eleven typical spatial hotspots of violent crimes were identified,and their distribution regularities are: violent crimes usually occur in the areas with densely population and large population flow,and the numbers of crimes are positively correlated with the population density.By risk factor mining method based on word cloud analysis,it is found that the important risk factors in the spatial hotspots of injury cases are "whether there is dispute" and "whether there is alcohol abuse",the temporal hotspot for such cases is from 18:00 to 23:00,and the bars are the high-risk places.The important risk factors in the spatial hotspots of rape cases are "whether there is drunkenness" and "whether there stays drunkenness".23:00 to 3:00 am is the temporal hotspot,the compartment,toilets,forests and by-places are the high-risk places,and young college students and unemployed vagrants are high risk groups related to the cases.In the spatial hotspots of robbery cases,21:00 to 2:00 am is the temporal hotspot,allies,subway entrances and bank surroundings are high risk places,and women,unemployed people,illiterates and migrant workers are high risk groups related to the cases.By analysis of the consequences of violent crimes based on machine learning,it is found that the performance of the prediction model of assault cases' consequences obtained by takingrandom forest algorithm is the optimal(the average value of F1 through 10-fold cross-validation is 0.72).It is explored that the risk factors such as "alcoholism","number of people related to cases","crime modus operandi" and "whether it is a hot spot time" excavated by using the feature importance ranking have a significant impact on the consequence severity of assault cases.The above method can be preferably applied in the violent crimes' spatial-temporal hotspots analysis and risks analysis,and it can accurately identifies the spatial-temporal hot spots of crimes,dig out risk factors,and predict the consequence severity,which can provide technical and decision support for public security departments to prevent and control violent crimes.