Accident Data Analysis And Road Safety Using Ai And Ml

Abstract

Road traffic accidents (RTAs) remain one of the leading causes of mortality and economic loss worldwide, necessitating intelligent data-driven approaches for prevention and mitigation. This paper proposes an AI-driven framework for accident data analysis and road safety enhancement that leverages machine learning (ML), feature engineering, and hierarchical clustering techniques to predict accident severity, identify key risk factors, and uncover spatial-temporal accident patterns. The proposed architecture comprises six integrated layers: data collection and ingestion, preprocessing, feature engineering, predictive modeling, clustering, and decision support. Using a comprehensive dataset of over 120,000 accident records from Victoria (2012–2023), three ML models — Logistic Regression, Random Forest, and XGBoost — were trained and evaluated. Among these, XGBoost achieved the highest performance with an accuracy of 91.2% and F1-score of 90.4%, outperforming baseline models. Feature importance analysis highlighted critical factors such as time of day, speed limit, lighting conditions, and road geometry, offering actionable insights for traffic authorities. Furthermore, hierarchical clustering identified high-risk zones and recurring accident patterns, enabling targeted safety interventions. The results demonstrate the potential of AI-based accident analysis systems to support data-driven policymaking, improve emergency response strategies, and contribute to the Vision Zero initiative aimed at eliminating traffic fatalities. This research lays the foundation for future intelligent transportation systems that integrate real-time sensor data and explainable AI for dynamic risk assessment and predictive road safety planning