Correlation analysis between urban environment features and crime occurrence based on explainable artificial intelligence techniques

Crime and urban environments are considered to be closely related. However, there exists no clear understanding of this phenomenon. Therefore, this study analyzes the relationship between various urban environmental variables and crime occurrences and provides insights into the optimal placement of crime prevention facilities by developing a crime prediction map based on a paradigm of the Daegu city. To achieve this, we used 373,387 crime reports from Daegu as dependent variables and 370,000 random points. Independent variables included information such as the point of interest, land use, land cover, floating population, and card sales. The developed crime prediction map created using the model was used to evaluate the adequacy of CCTV installation locations and identify areas requiring new CCTV installations. The performances of various machine-learning models were compared and the XGBoost model (accuracy of 89.7 % and precision of 89.8 %) was selected. Key variables influencing crime report data were identified using the SHAP(SHapley Additive explanation) method. To analyze the spatial explanatory power of the relationship between crime and urban environmental variables, various buffer distances were tested, and a 20 m buffer distance was derived. The results of this study are expected to provide valuable data for crime prevention policies.