Unveiling pedestrian injury severity risk factors: A comparative analysis of machine learning and deep learning methods

• The study investigates the risk factors in vehicle crashes involving pedestrians is vital to developing and implementing effective countermeasures and policies. • In this study two machine learning models (Random Forest and Gradient-Boosted Trees) and a Deep Neural Network (DNN) model are employed. • Further, a comprehensive analysis of vehicle-hit pedestrian crashes is conducted using the Shapley Additive exPlanations (SHAP) analytics. • Modeling results revealed that the DNN model outperformed other models, achieving a prediction accuracy of 93.51%. Pedestrian safety poses a pressing public health concern globally, accounting for a significant proportion of traffic-related injuries. Understanding the risk factors in vehicle crashes involving pedestrians is vital to developing and implementing effective countermeasures and policies. Existing studies have mainly focused on identifying pedestrians’ risk predictors through conventional statistical and machine learning models. Despite the growing importance of these models in traffic safety, the consistency of risk factors among machine learning and deep learning approaches has been scarcely explored. This study fills a research gap by modeling pedestrian traffic collisions through systematic machine and deep learning frameworks and comparing the consistency of severity risk factors between both methods. Pedestrian crash data from Rawalpindi, Pakistan, are used to develop two machine learning models (Random Forest and Gradient-Boosted Trees) and a Deep Neural Network (DNN) model. Further, a comprehensive analysis of vehicle-hit pedestrian crashes is conducted using the Shapley Additive exPlanations (SHAP) analytics. Modeling results revealed that the DNN model outperformed other models, achieving a prediction accuracy of 93.51%. Analysis for critical risk factors using the Random Forest model indicates that variables such as season of the year, weather types, and crash characteristics (such as overspeeding, distractions, and vehicle-pedestrian collision types), time of the crash are consistent with those obtained from DNN model. These consistent results highlight the ability of machine learning and deep learning models to provide interpretable insights that can direct the development of affordable and effective safety measures to enhance pedestrian safety