Abstract: This research introduces a hybrid framework combining ensemble-based learning and anomaly detection for the prediction of road accident severity. Conventional predictive systems often fail to manage noisy and imbalanced accident datasets effectively. To address this limitation, the proposed design integrates Decision Tree and Random Forest classifiers with clustering methods—KMeans and DBSCAN—for simultaneous classification and hotspot detection.
Contextual factors such as weather patterns, road conditions, traffic intensity, and casualty ratios are incorporated through tailored feature engineering. The Random Forest model achieved an accuracy of 83.6%, surpassing baseline methods. By fusing anomaly detection with ensemble classification, the framework not only enhances prediction accuracy but also provides interpretable insights for preventive traffic management and policy-making.

Keywords: Road Accident Severity · Anomaly Detection · Ensemble Models · Random Forest · Traffic Safety · Machine Learning.

Downloads: | DOI: 10.17148/IJIREEICE.2025.131125