<p>Accident severity prediction is a critical challenge in traffic safety management, emergency response, and urban mobility planning. Road accidents remain a leading cause of fatalities worldwide, yet existing accident analysis frameworks often lack predictive accuracy, interpretability, and real-time decision-making capabilities. Traditional statistical models fail to capture complex interactions between vehicle attributes, driver behavior, and environmental factors, limiting their effectiveness in accident severity assessment. This study addresses these gaps by developing an explainable artificial intelligence (XAI) framework for accident severity prediction, leveraging machine learning models (Random Forest, Extreme Gradient Boosting, Support Vector Machine, and Naïve Bayes) and SHapley Additive exPlanations (SHAP) analysis to enhance model transparency. The dataset, sourced from the Fatality Analysis Reporting System (FARS), consists of 7394 recorded crashes and incorporates key predictors such as airbag deployment, control devices, seatbelt usage, and driver demographics. Experimental results demonstrate that XGBoost outperforms other models, achieving the highest accuracy (80.8%), recall (80.8%), and F1-score (81.0%), making it the most reliable classifier for distinguishing between severe and non-severe accidents. SHAP analysis reveals that airbag deployment, seatbelt usage, and control devices significantly impact accident severity outcomes, providing valuable insights into policy-driven interventions and traffic management strategies. Despite its effectiveness, the study highlights limitations such as data imbalance, lack of real-time behavioral factors, and exclusion of non-fatal crashes, suggesting deep learning integration, real-time telematics, and hybrid AI models in future research. The proposed framework offers a data-driven approach to accident severity prediction, improving road safety policies, vehicle design enhancements, and emergency response efficiency.</p>