Modeling Pedestrian Priority at Unsignalized Crossings Using Machine Learning: Toward Data-Driven Transport Policy Solutions

Pedestrian safety at unsignalized crossings remains a pressing urban mobility challenge, where priority negotiation between pedestrians and vehicles directly influences crash risk and traffic efficiency. This study presents a data-driven framework employing machine learning to predict pedestrian priority, integrating behavioral, temporal, vehicular, and contextual features while accounting for spatial heterogeneity via Location ID and addressing class imbalance in each location using SMOTENC oversampling. Three models, Random Forest, XGBoost, and Support Vector Machines, were trained and tested on behavioral, temporal, and contextual features. The XGBoost model achieved the highest predictive performance with a test accuracy of 0.78 (95% CI: 0.73–0.82). Feature importance analysis identified pedestrian behavior, waiting position, free time, and vehicle standing time as dominant predictors, with Location ID contributing meaningfully to capture spatial variation. In contrast, demographic factors such as age and sex exhibited negligible predictive value. These results underscore that temporal-behavioral dynamics and situational factors primarily govern priority negotiation, underscoring the importance of behaviorally responsive safety measures. The results provide direct implications for urban transport policy, including adaptive signage, crossing design modifications, and targeted interventions to encourage safe pedestrian behavior. Overall, the study demonstrates the dual methodological and practical contributions of machine learning in transport safety: enhancing predictive modeling of pedestrian–vehicle interactions and translating empirical evidence into actionable strategies for safer, more equitable urban crossings.