Explanatory Modeling of Accident Severity at Signalized Intersections: A Case Study of Manama, Bahrain

Understanding the factors associated with accident severity at signalized intersections is essential for developing context-specific safety interventions in Bahrain. This study applies explanatory modeling to investigate severe accident out-comes at urban signalized intersections in Manama region using accident-level data from 2013-2019. Binary severity models were developed using stepwise logistic regression, decision tree, and XGBoost to examine the influence of traffic exposure, inter-section geometry, temporal conditions, and selected behavioral categories. Model performance was evaluated using ROC-AUC and balanced accuracy to allow consistent comparison across approaches. Although all models demonstrated limited predictive capability, with performance only marginally above random classification, the explanatory analyses revealed consistent patterns. Traffic exposure measures, intersection density, and nighttime conditions emerged as influential factors in severity outcomes, while behavioral variables contributed minimally. These findings emphasize the importance of intersection design, traffic conditions, and operational context in accident severity at signalized intersections. The results also highlight the need for richer, locally relevant data to support improved explanatory and predictive safety modeling in Bahrain.