Analysis of Pedestrian- and Bicyclist-Involved Traffic Crashes Using Machine Learning

Equitable and sustainable urban mobility depends on reliable public transportation coupled with safe first- and last-mile travel. However, the high rate of fatal and severe injury crashes involving nonmotorists poses a major barrier to reducing automobile dependency. Although numerous studies have examined crash severity, few have focused specifically on vulnerable road users using interpretable, comparative machine learning approaches. This case study applied four machine learning models—support vector machine (SVM), random forest, k-nearest neighbors (KNN), and artificial neural network (ANN), to analyze over a decade of crash data from Franklin County, Ohio. This research focused exclusively on nonmotorist crashes and combined predictive modeling with SHapley Additive exPlanations (SHAP) and partial dependence plot (PDP) analysis to identify the most influential risk factors in an interpretable manner. The ANN model achieved the highest area under the curve (AUC) performance and was used to identify key contributors to severe and fatal crashes, including lighting conditions, turning movements, substance use, and traffic control. By integrating comparative modeling with interpretable AI techniques, this study offers a framework for understanding and mitigating nonmotorist crash risks, advancing data-driven safety planning for vulnerable road users in urban environment.