Crime mapping in urban environments using explainable AI: A case study of Daegu, Korea

초록

Urban safety constitutes a foundational pillar for sustainable cities. This study aims to analyze the relationship between various crime types and urban environmental characteristics in the case of Daegu Metropolitan City, South Korea, and to develop a predictive crime risk map. Spatial data were constructed by generating buffer zones ranging from 10 to 50 m around 112 emergency call locations. Utilizing the XGBoost algorithm in conjunction with SHAP (SHapley Additive exPlanations), we identified key variables influencing different types of crime. The analysis revealed that commercial facilities, floating population, credit card transaction volumes, number of households, and road area were consistently significant predictors across all crime categories, with the highest predictive performance observed at the 10-meter buffer scale. Based on the prediction results, we developed high-resolution crime probability maps that visualize high-risk areas and assessed the adequacy of existing CCTV installations. By providing interpretable, spatially detailed, and high-resolution forecasts, this research enables data-driven crime prevention strategies and supports sustainable urban planning and inclusive safety policies. The proposed approach is inherently scalable and transferable to a variety of urban contexts that encounter difficulties in synchronizing safety infrastructure with rapidly evolving urban dynamics.

키워드