Car accidents remain a major global concern, with accident severity influenced by a complex interplay of factors such as road conditions, weather, and driver behaviour. Predicting accident severity through automated methods of classification and forecasting presents significant challenges, requiring advanced analytical techniques and robust algorithms. This paper presents a comprehensive systematic review of methodologies proposed in the literature that apply Artificial Intelligence (AI) and Machine Learning (ML) approaches to predict traffic accident severity and its contributing factors. The review critically examines the effectiveness of various AI and ML models, comparing traditional statistical approaches with advanced deep learning and hybrid techniques. A comparative analysis of methodologies is conducted to identify common patterns, methodological gaps, and key strengths across the reviewed studies. The findings highlight the growing importance of real-time data integration, predictive modelling, and ensemble learning techniques in improving accuracy and advancing road safety applications. Furthermore, several studies report exceptionally high predictive performance, with models such as XGBoost combined with another deep learning algorithms achieving accuracies of up to 98%, demonstrating the superiority of advanced AI-based frameworks over conventional ML models.
Alobidan et al. (Fri,) studied this question.