ACCIDENT SEVERITY DETECTION BY USING MACHINE LEARNING

Road accidents cause significant human and economic losses, making severity prediction an important research area. This project presents a machine learning–based approach to classify road accidents into three categories: Fatal, Serious, and Slight. The dataset includes important attributes such as road surface conditions, weather conditions, lighting conditions, vehicle types, and casualty information. These features are analyzed to identify patterns that influence accident outcomes.

Multiple models are implemented and compared to determine the most effective approach. Traditional algorithms such as Logistic Regression and Random Forest are used as baseline models. Advanced boosting techniques including XGBoost, LightGBM, and CatBoost are applied to improve prediction capability. In addition, a Feedforward Neural Network is developed to capture nonlinear relationships among features. Ensemble methods such as Stacking and Voting classifiers are also used to enhance overall performance.

The models are evaluated using accuracy, precision, recall, F1-score, and ROC–AUC metrics. Class imbalance handling techniques are applied to ensure fair classification across all severity levels. The results show that boosting and ensemble models provide improved predictive performance. The proposed framework supports data-driven analysis and contributes to improved safety planning and decision-making.

Keywords: Machine Learning, Road Accident, Severity Prediction, Random Forest, XGBoost, LightGBM, CatBoost, Neural Network, Stacking, ROC–AUC.