Road Traffic , Accident Risk and Severity Prediction System Using Hybrid Machine Learning Models

This research presents a hybrid machine learning-based system for predicting road traffic accident risk and severity. The system integrates multiple models including Random Forest, XGBoost, LightGBM, and CatBoost using a majority voting approach to improve prediction accuracy. The model incorporates environmental, traffic, and geographical data along with feature engineering techniques such as time-based features and geo-clustering. A dynamic risk score is calculated based on visibility, traffic density, and weather conditions. The system provides outputs including accident risk level, severity classification, and visualization of accident-prone areas using heatmaps. Experimental results demonstrate that the hybrid model outperforms individual models in terms of accuracy and reliability. This work contributes to intelligent transportation systems by enabling proactive decision-making and improving road safety.