Intelligent Accident Severity Prediction for Faster Emergency Response using RF-RFE and Deep Learning Model

B.Yamini Priyanka1 ,Lakkireddy Vyshnavi2 ,Patnam Hanesha3 ,Sontam Lokeshwar Reddy4 , MoturuMadhura5
1Assistant Professor, Dept of Information Technology,SV College of Engineering, Tirupati,India.
2B.Tech, Dept of Information Technology,SV College of Engineering, Tirupati,India.
3B.Tech, Dept of Information Technology,SV College of Engineering, Tirupati,India.
4B.Tech, Dept of Information Technology,SV College of Engineering, Tirupati,India.
5B.Tech, Dept of Information Technology,SV College of Engineering, Tirupati,India.

Abstract - Accident severity refers to the classification of road traffic accidents based on the level of harm caused,often categorized as fatal, serious, minor injuries, or no injury. Severity prediction systems use factors like vehicle type,driver behavior, weather, lighting, and traffic conditions to classify and predict the outcome of accidents. Accurate severity prediction aids in prioritizing emergency response, improving resource allocation, and enhancing overall road safety management by enabling faster and more informed decisions. The existing system predicts road accident severity using machine learning and deep learning to enhance emergency response. It applies Random Forest Recursive Feature Elimination (RF-RFE) for optimal feature selection and SMOTE-Tomek for data balancing. A hybrid CNN-BiLSTMAttention model captures spatial, sequential, and critical patterns, while SHAP provides interpretability by identifying key severity factors. Evaluated on a French accident dataset, the system demonstrates high accuracy and reliability. The proposed next-generation emergency response system enhances the current model by integrating real-time, multi-source data such as environmental conditions, traffic flow, and temporal factors. Using spatiotemporal modeling and causal inference, it captures dynamic severity patterns with greater accuracy across urban and ural areas. Linked with live traffic and dispatch systems, it enables real-time prioritization and optimal resource allocation. Micro-level SHAP analysis ensures deeper insights for policymakers, improving prediction accuracy, response efficiency, and overall roadsafety management.Keywords: Random Forest Recursive Feature Elimination (RF-RFE), machine learning, deep learning, SMOTE Tomek, CNN-BiLSTM-Attention model, Micro-level SHAP, spatiotemporal modeling.