The Urban Intersection Accident Detection Method Based on the GAN-XGBoost and Shapley Additive Explanations Hybrid Model

Traffic accidents at urban intersections may lead to severe traffic congestion, necessitating effective detection and timely intervention. To achieve real-time traffic accident monitoring at intersections more effectively, this paper proposes an urban road intersection accident detection method based on Generative Adversarial Networks (GANs), Extreme Gradient Boosting (XGBoost), and the SHAP interpretability framework. Data extraction and processing methods are described, and a brief analysis of accident impact features is provided. To address the issue of data imbalance, GAN is used to generate synthetic accident samples. The XGBoost model is then trained on the balanced dataset, and its accident detection performance is validated. In addition, SHAP is employed to interpret the results and analyze the importance of individual features. The results indicate that the accident samples generated by GAN not only retain the characteristics of real data but also enhance sample diversity, improving the AUC value of the XGBoost model by 7.1% to reach 0.844. Compared with the benchmark models mentioned in the study, the AUC value shows an average improvement of 7%. Additionally, the SHAP model confirms that the time–vehicle ratio and average speed are key factors influencing the model’s detection results. These findings provide a reliable method for urban road intersection accident detection, and accurate accident location detection can assist urban planners in formulating comprehensive emergency management strategies for intersections, ensuring the sustainable operation of traffic flow.