Modeling Of Spatial Distribution Density And Influential Factors Of Urban Crimes Based On Random Forests

Since the Reform and Opening-up policy,the process of urbanization in China has been accelerated rapidly,especially in large and medium-sized cities.During the process of rapid development,the urban crimes has become one of the main stumbling blocks to the sustainable urban development.Hence,decreasing criminal activities effectively is an unavoidable challenge against the background of rapid urbanization.Based on the existing studies related to the modeling of urban crimes and analysis of influential factors,this research analyzed the spatio-temporal distribution of six typical crimes,including burglary,outdoor theft,robbery and snatching,pornographic activities,insult and assault,and violent crimes in Shanghai in the year of 2015.Multi-source data including landuse types,road network,Point Of Interest(POI),nighttime light images,population,and housing price were then integrated into a set of factors based on the 500m grid.Furthermore,a quantitative relationship was established between the crime intensity and various factors by using the Random Forest algorithm,and the effects of different factors on the occurrence of crimes were analyzed.The major achievements are as follows:(1)The distribution of each criminal activity was analyzed from time and space perspective according to the report records.The temporal analysis indicated that there were obvious seasonal and day-night variation among criminal activities.Comparative analysis at different spatial scales demonstrated that the criminal activities were concentrated in the urban center area(inside the outer ring).(2)The spatial kernel density of criminal cases under the grid scale of 500m was used as an indicator of crime intensity in this research.Then,with the support of Random Forest algorithm,the prediction model of criminal intensity was established within the outer ring and extracted the important factors.The evaluation parameter~2of six models were all around 0.80 and the Root Mean Square Errorwas low.And the test results also showed that the models had better prediction ability.(3)On the basis of the constructed prediction models of crime intensity,the feature contribution was introduced to explorer the relationship between urban characteristics and crime intensity.The results indicated that the influence of a factor may vary across different crime types and there was also obvious differences in influence among factors related to the same crime type.The result of this research is valuable in that it can provide decision-making support for governance,prevention,and control of urban crimes.Moreover,the proposed research framework based on Random Forests provides a new perspective for the future study of urban crimes that may employ the algorithm.