Artificial intelligence in traffic accident analysis: A comprehensive review
ADVANCED ENGINEERING INFORMATICS, cilt.77, 2027 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Derleme
- Cilt numarası: 77
- Basım Tarihi: 2027
- Doi Numarası: 10.1016/j.aei.2026.105202
- Dergi Adı: ADVANCED ENGINEERING INFORMATICS
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Applied Science & Technology Source, Compendex, INSPEC, Academic Search Ultimate (EBSCO), Engineering Source (EBSCO)
- Recep Tayyip Erdoğan Üniversitesi Adresli: Evet
Özet
Road traffic accidents continue to be a significant problem in terms of public health and transport safety. Whilst traditional statistical methods are of fundamental importance for drawing conclusions and evaluating policies, certain approaches have limitations stemming from predefined functional structures, distribution assumptions, limited forms of interaction, and difficulties in processing heterogeneous, high-dimensional or continuously updated data. This review critically synthesises artificial intelligence (AI) applications that directly support safety decisions through accident detection and prediction, the determination of accident frequency and risk, and the assessment of injury severity and duration. The literature is organised within a two-dimensional framework. In the first dimension, studies are classified as vision-based approaches, data-driven AI techniques and spatiotemporal AI models. The second dimension focuses on the roles of these studies in detection, prediction, explanation and decision support. The framework highlights developments ranging from standalone machine learning, convolutional and recurrent models to multimodal fusion, graph-based learning, transformers, foundation models, generative AI and digital-twin-supported simulations. However, the increase in predictive power gives rise to various trade-offs in terms of interpretability, robustness, computational cost, scalability, transferability, data requirements, fairness and readiness for deployment. Consequently, this study focuses not only on performance but also on the explanatory capacity of model families, their applicability and their contribution to engineering and policy decisions. Cross-regional comparison, causal and uncertainty-aware modelling, rare event learning, vision-language reasoning, synthetic scenario generation, digital twin validation and human-supervised decision support systems have been identified as key research priorities.