Spatiotemporal and situational analysis of motorcycle theft in Medellín, Colombia

Abstract Motorcycle theft is a major urban security problem in Medellín, Colombia, where motorcycles are central to mobility and income generation. This study examined motorcycle theft at two analytical levels. At the meso level, SAC/SARAR spatial regression models identified spatiotemporal clustering and area-level correlates of theft rates. At the micro level, a matched case-control analysis of CCTV footage compared stolen motorcycles with comparable non-targeted motorcycles in the same event settings; XGBoost and SVM classifiers assessed the predictive value of situational features. Theft clustered across time and space and was associated with commercial activity, road configuration, and surveillance-related features. Motorcycle cover, escape-route saliency, chromatic contrast, and motorcycle size emerged as the most informative predictors.