Abstract
This study introduces a methodological framework for predicting road accident severity using a SHAP-enhanced Machine Learning model. Road traffic accidents remain a major global concern, with India reporting over 150,000 fatalities annually. Traditional models fail to capture the complex relationships among various risk factors. This research applies machine learning, specifically Random Forest and Gradient Boosting, to identify and analyse key factors influencing accident severity. SHAP values are used to enhance model interpretability, providing insights into the contribution of each feature.•Develop a Random Forest model and a Gradient Boosting model to predict road accident severity based on a comprehensive set of features.•Utilise SHAP to identify and rank the importance of features, such as vehicle type, weather, and road conditions.•Model performance is evaluated using accuracy, precision, recall, F1-score, and confusion matrices. Polynomial curve fits are used only as post-hoc visualizations of the Actual-Predicted relationship (on ordinal codes), not as classifier evaluation metrics.The findings highlight that factors like vehicle type, accident location, and road conditions significantly influence accident severity. This approach provides a scalable and interpretable framework for improving road safety on Indian highways, offering data-driven insights for proactive safety measures and infrastructure enhancements.