Understanding Risk Factors for Pedestrian Injury Severity in Palermo, Italy: A Data-Driven Safety Perspective

Pedestrian injuries remain a major public health challenge in urban areas worldwide, especially in cities where dense populations, aging infrastructure, and complex street layouts create hazardous conditions for vulnerable road users. Despite growing attention to pedestrian safety, there is still limited evidence on how specific combinations of environmental, behavioural, and temporal factors are associated with pedestrian injury severity in cities with historically constrained urban environments. This study addresses this gap by analysing pedestrian crash severity in Palermo, Italy, a city characterized by narrow streets, high traffic volumes, and unique urban design challenges. Drawing on eight years of crash data, the research identifies the most influential factors contributing to the prediction of pedestrian injury severity. To address incomplete and unevenly reported crash records, the study implements a multifaceted imputation strategy to enhance data reliability and strengthen the consistency of severity-prediction models. In addition, machine learning models, including CatBoost, Random Forest, XGBoost, and Gradient Boosting, were trained for severity prediction, and global/local SHAP analyses and interaction effect visualizations were used to support the interpretation of model outputs. The findings indicate that male pedestrians, two-way roads, straight sections, poor lighting, nighttime periods, and rainy weather are significantly associated with a higher predicted likelihood of severe injuries. The interaction analysis further illustrated that higher predicted severity frequently clusters in specific combinations, including male pedestrian cases occurring on two-way/straight segments under nighttime/low visibility and rainy weather contexts. In addition, the analysis highlighted that pedestrian crashes on crosswalks were predicted to be more severe during weekdays, while nighttime crashes were predicted to be more severe on weekends. The findings support data-informed prioritization of safety evaluations and targeted countermeasure assessment in historically constrained city centres.