Predicting road traffic crash severity using machine learning models: A comparative study

Road traffic crashes remain a major safety challenge in rapidly urbanizing cities, often resulting in serious injuries or fatalities. This study evaluates four machine learning (ML) models-Extreme Gradient Boosting (XGBoost), Random Forest (RF), Support Vector Machine (SVM), and Convolutional Neural Network (CNN)-for classifying crash severity using Addis Ababa crash records from 2017 to 2020. Crash severity was grouped into low (property damage only), medium (injury), and high (fatal) classes. The models were trained and optimized using a transparent comparative framework with preprocessing, class balancing, and explainability analysis. XGBoost achieved the strongest overall performance, followed by Random Forest, while SVM and CNN showed lower classification capability. Shapley Additive Explanations (SHAP) indicated that environmental and behavioral factors, including road division, junction type, and risky maneuvers, had greater influence than demographic variables. The study provides a transparent and interpretable ML framework for crash severity classification in a developing urban context.