INTEGRATING SITUATIONAL FACTORS AND HETEROGENEOUS DATA FOR ENHANCED TRAFFIC ACCIDENT ANALYSIS

Abstract

Road traffic accidents (RTAs) pose a significant threat to public safety and transportation networks, requiring advanced modeling techniques to study their causes, trends, and consequences. The complexity of RTAs modeling stems from the need to integrate diverse data sources, including spatiotemporal changes, environmental variables, vehicle and driver characteristics, and infrastructure elements. Recent advances in machine learning, including graph neural networks, attention-based deep learning models, and hybrid tree classifiers, have improved prediction accuracy and interpretability. These methods leverage heterogeneous data to provide real-time crash prediction, severity assessment, and identification of problem areas. However, obstacles remain, such as data sparsity, class imbalance, and model interpretability. This article reviews modern approaches to RTAs modeling and identifies the impact of diverse information on crash frequency and severity estimates, thereby improving road safety, optimizing traffic management, and refining crash prevention approaches and technologies. Promising research directions in this area of modeling and machine learning are identified: integration of heterogeneous data, use of causal machine learning methods and real-time decision support systems.