Predicting Traffic Crash Severity in Jeddah Using Machine Learning Techniques: A Data-Driven Approach to Road Safety

Abstract

Road traffic accidents (RTAs) remain a significant public safety concern, particularly in rapidly urbanizing regions such as Jeddah in Saudi Arabia. This study investigates the prediction of RTA severity using a data set of 877 RTAs and 22 independent variables, including driver characteristics, road and weather conditions, and traffic violations. Advanced machine learning (ML) models—decision tree, random forest, LightGBM, XGBoost, CatBoost, and AdaBoost—were used to predict crash outcomes. The XGBoost model demonstrated the highest predictive performance, with an accuracy of 85%, precision of 83%, recall of 84%, F1-score of 84%, and a receiver operating characteristic (ROC) area under the curve (AUC) of 0.92. Shapley additive explanations (SHAP) analysis revealed that traffic violations, excessive speed, and adverse weather conditions were the most influential factors in determining crash severity. These findings provide actionable insights for policymakers, emphasizing the importance of enforcing traffic regulations, addressing infrastructure deficiencies, and mitigating weather-related risks. By using ML insights, this study supports efforts to enhance road safety and promote sustainable urban development in Jeddah.