Multidimensional analysis of e-scooter crash severity using cluster correspondence and explainable AI

The rapid expansion of electric scooters (e-scooters) in urban transportation networks has transformed personal mobility across U.S. cities, yet rising crash rates have raised critical safety concerns. This study investigates the factors contributing to e-scooter crash severity through a novel integration of Cluster Correspondence Analysis (CCA) and SHapley Additive exPlanations (SHAP), offering both structural and interpretable insights into injury outcomes. Using 355 e-scooter crash records from Texas between 2021 and 2024, the study identifies four distinct crash typologies based on behavioral, spatial, and environmental features. These clusters capture key contexts including intersection-related crashes with traffic control presence, midblock conflicts in shared lanes, incidents in informal low-speed areas like parking lots, and crashes at driveways and off-road access points. SHAP analysis quantifies the contribution of individual predictors to severity outcomes, revealing that intersection complexity, lighting condition, road classification, rider age, and crash configuration (such as direction of impact or off-road transitions) are consistently associated with severe injuries. Higher severity was observed among older riders, in environments with poor visibility, and at points of vehicle-e-scooter interaction such as signalized intersections and driveway exits. To support applied use, an interactive Shiny dashboard was developed to visualize crash patterns and assist in the identification of high-risk locations. The findings underscore the importance of context-specific factors in e-scooter crash outcomes and offer actionable insights for improving micromobility safety through targeted infrastructure design, regulatory interventions, and rider-focused educational strategies.