| Abstract: |
Road traffic accidents represent one of the most critical public safety challenges worldwide, causing millions of fatalities and injuries annually. The increasing availability of large-scale accident datasets, combined with rapid advances in machine learning (ML) and artificial intelligence (AI), has catalyzed a surge of research efforts aimed at understanding, predicting, and mitigating road accident occurrences. This review paper presents a systematic and comprehensive meta-analysis of past work on road accident analysis using machine learning techniques, synthesizing findings from over three decades of published literature across diverse geographic, environmental, and technical contexts. The paper examines the evolution of ML approaches applied to accident prediction, severity classification, hotspot detection, causal factor identification, and post-accident response optimization. Key algorithms reviewed include decision trees, random forests, support vector machines, artificial neural networks, gradient boosting methods, deep learning architectures, and ensemble models. The study critically evaluates dataset sources, feature engineering strategies, model evaluation metrics, and the translational impact of findings on traffic policy. Findings reveal that ensemble methods and deep learning models consistently outperform traditional statistical approaches, achieving accuracy rates of 80 to 96 percent across various sub-tasks. However, significant gaps remain in model interpretability, real-time deployment, cross-regional generalizability, and integration with intelligent transportation systems. This paper provides a structured synthesis of existing knowledge and highlights priority research directions for future work in ML-driven road safety analysis. |
