| Crime can not only harm the physical and mental health of victims directly,but can also affect the harmony and stability of society.Therefore,it is necessary to deepen the understanding of crime for the formulation of effective prevention and control strategies.As a kind of geographical event,crimes occur in a certain space,so space is very important for crime analysis.The advancement of policing techniques and the development of geographic information technology and computer technology have created necessary conditions for the study of crime from the geospatial perspective.In recent years,more and more attention has been paid to the study of crime geography,especially on the micro spatial level.However,most of the existing studies are based on a single spatial level,and the results have some limitations,which can not provide a comprehensive understanding of criminal cases.In this context,this thesis reviews the development history of crime geography and summarizes the international and domestic research status.Based on relevant theories of crime geography,and takes residential burglary in Hankou area of Wuhan City as the experimental object,this thesis uses spatial analysis and spatial statistical techniques to study how spatial distribution,influencing factors,and space-time hotspots prediction performance of residential burglary are affected by the multi-level space structure.This study enriches the research of crime geography to a certain extent,and the relevant results can provide a reference for the analysis work of scientific researchers and the crime prevention and control decision-making of the police department.Specifically,the research work of this thesis mainly includes the following aspects:(1)Space-time exploratory data analysis of residential burglaryIn order to get a preliminary understanding of the spatial and temporal distribution characteristics of residential burglary,this study first conducts exploratory data analysis.Considering the three different spatial levels(i.e.,jiedao,community,and street-section)in the study area,the hot spot distribution characteristics of residential burglary on each space level are analyzed by planar and network kernel density estimation methods;then,the spatial autocorrelation patterns at different space levels are examined by global and local Moran’s index from global and local perspective,respectively.Finally,the numbers of residential burglaries within different time periods(including day,week,and hour)are analyzed.The results show that residential burglary cases show significant positive autocorrelation at three spatial levels,and the autocorrelation at street-section level is the largest,followed by community and jiedao level;the time exploratory data analysis results show that the number of daily cases fluctuates greatly,the number at weekends is less than the working days,and the number of criminal cases is the largest in the early peak period.(2)Analysis of aggregation patterns of residential burglary at different space levelsIn order to understand how the spatial aggregation pattern of residential burglary is affected by the layout of different levels of space,this study analyzes the spatial distribution pattern of residential burglary from three space levels based on the multilevel structure of geographical space.First,the Lorenz curve and the Gini coefficient method are used to examine the overall aggregation degree of residential burglary at three space levels;then,considering the nesting structure of space units on different levels,we use multi-level Gini coefficient to analyze how the multi-level spatial layout influences the aggregation degree.Finally,the group trajectory model is used to analyze the differences in the evolution trends of the number of cases with the change of time.The results show that residential burglary cases show significant aggregation at three space levels and the largest aggregation degree is at the street-segment level,that is,the lower the space level,the greater the aggregation degree of crime cases;the distribution pattern of crime cases at the lower space level will be affected by the higher space structure,that is,different space levels have an impact on the space distribution of crime cases;most of the street-sections in the region show a stable and low crime rate,while a large proportion of cases are gathered in a small part of the street-sections,and these street-sections also show the greatest downward trend.(3)The impacts of residential burglary from different levels and the interaction effects of cross-level impactsIn order to analyze the causes of residential burglary,this thesis considers the influences from different space levels and the interaction of cross-level features.First of all,we use the unconditional multi-level regression model to examine whether the distribution of residential burglary is affected by the multi-level structure of space.Then,based on existing research,street-level features(such as road network structure)and community-level features(such as population,housing characteristics,land use characteristics,etc.)that may have an impact on the formation of residential burglary are selected.Finally,considering the overdispersion characteristics and spatial autocorrelation effect,four kinds of negative binomial multi-level regression models are constructed,which includes street level features,community-level features,twolevel features,two-level features and cross-level features respectively.Based on AIC,BIC,and log-likelihood values,the fitting effects of the four models are compared and analyzed in detail.The results show that the calculation method of street permeability proposed in this study is more effective than the traditional method;street permeability has a significant impact on residential burglary,and the population density,road density,average house age and other characteristics at the community level have a significant positive correlation with burglary;there is interaction between the characteristics at different levels,that is,the characteristics at the community level will impact the influence of the characteristics of its internal street-sections on burglary cases.(4)The effect of space level and different parameters on the prediction performance of space-time hot spotsIn order to predict the space-time hot spot of residential burglary,this study extends the common planar kernel density estimation method to three-dimensional space,and analyzes the influence of space level and various parameters on the prediction effect.Firstly,the principle of space-time kernel density estimation and the parameters that need to be specified when constructing the space-time kernel density estimation model are introduced in detail,including the size of the spatiotemporal unit,the space bandwidth and the time bandwidth;secondly,the commonly used parameter selection methods are introduced,and a data-driven optimal space-time bandwidth selection method is introduced;finally,based on different parameters,the space-time kernel density estimation model is constructed to predict the space-time hot spots of residential burglary.Hit rate and prediction accuracy index are used to compare and analyze the performances of different models.The results show that the spatiotemporal kernel density estimation model at street network level has higher prediction accuracy than the spatiotemporal kernel density estimation model at the plane level,and the influence of the size of spatiotemporal unit on the prediction effect of spatiotemporal kernel density estimation model is limited,while the influence of spatiotemporal bandwidth is greater,especially the spatial bandwidth. |
PDF Full Text Request