Authors: Research Scholar Mitali Khandelwal, Dr. Kshmasheel Mishra
Abstract: Road traffic accidents are a leading cause of fatalities and injuries worldwide, with their severity influenced by numerous dynamic and interrelated factors. Predicting the severity of such accidents is critical for timely emergency response, infrastructure planning, and overall road safety enhancement. This study explores advanced machine learning (ML) techniques to accurately predict accident severity using diverse datasets comprising weather conditions, road characteristics, vehicle information, and driver behavior. A novel hybrid model is proposed, integrating classical pre-trained models with ensemble learning and deep neural networks to capture complex nonlinear relationships within the data. To address class imbalance a common issue in accident datasets, strategies such as data augmentation and cost-sensitive learning are incorporated. The proposed architecture undergoes rigorous performance evaluation using metrics like accuracy, precision, recall, and F1-score, demonstrating superior predictive capability compared to traditional models. Further, the study emphasizes model interpretability through tools like SHAP and LIME to enhance transparency and trust. By leveraging real-world traffic data and scalable ML infrastructure, the developed model offers a reliable solution for severity prediction, aiding emergency services and policy decisions. Ultimately, this research contributes to the reduction of road casualties and the advancement of intelligent transport systems.
