Spatial-Temporal Analysis and Severity Prediction of Traffic Accidents
Abstract
Traffic accidents pose a significant problem, resulting in substantial human and economic losses annually. As such, safety engineers rely on studies of traffic accidents to implement remedial measures that prevent accidents or mitigate their impact on lives and economy. This study explored spatial and temporal traffic accident patterns and developed a data-driven traffic accident severity prediction model. The traffic accident data used in the study were acquired for the City of Irbid, which is located in Jordan, covering the period from January 2016 to December 2021. The accident severity prediction model was developed using a Random Forest machine learning classification algorithm, and its optimal architecture was identified using a Bayesian optimization approach. Additionally, a model interpretability method, SHapley Additive exPlanations (SHAP), was employed to examine the impacts of the explanatory variables and their relative importance. The resulting model showed outstanding performance, with an overall predictive accuracy of 90%. Furthermore, the SHAP analysis results revealed that accident speed had the greatest influence on the model’s predictions, followed in order by accident type, date, and primary road class. The results also indicated that higher accident speed values, collisions, and Run-Off-Road accident types, and downgrade curves are associated with higher accident severity classes.