Exploring Accident Frequency Using Explainable AI: a Case Study of the Central Section of National Freeway No.1 in Taiwan

In recent years, with the widespread adoption of artificial intelligence (AI) technology, numerous studies have employed AI for traffic accident analysis and prediction, replacing traditional statistical methods. In Taiwan, traffic accidents occur frequently on a daily basis, making it crucial to investigate the factors contributing to their occurrence. This study utilizes SHapley Additive exPlanations (SHAP), an explainable AI (XAI) technique, in conjunction with two regression models—linear regression and random forest regression—to examine the impact of freeway environmental factors on accident frequency. The study focuses on the central section of National Freeway No. 1 in Taiwan, using road environment variables provided by the Taiwan Ministry of Transportation and Communications and open data on vehicle detection traffic flow from 2021 to 2024. A total of 12 variables, including curve radius and the number of lanes, were analyzed. The SHAP method was used to evaluate feature importance and generate bee swarm plots for visualization. The findings from both regression models indicate that traffic volume and radius of curvature (A-value) are among the top three most influential environmental factors affecting accident frequency. These findings offer valuable empirical insights for enhancing freeway safety management.