Predicting Accident Severity on Taiwan Highways Using Machine Learning and Electronic Toll Collection (ETC) Data

This study aims to develop a machine learning‐based framework for predicting the severity of highway traffic accidents by leveraging high‐resolution data from Taiwan’s Electronic Toll Collection (ETC) system. Unlike traditional accident‐reporting systems, the ETC infrastructure provides a uniquely comprehensive and precise dataset that captures spatiotemporal traffic patterns and environmental conditions across the national highway network. This rich dataset enabled the integration of data mining and data visualization techniques to uncover nontypical contributing factors to accident severity. Feature engineering was conducted using random forest and LASSO regression, while extreme gradient boosting and the Apriori algorithm were employed to identify key associations between accident severity and contextual variables. Based on human factor and traffic psychology theory, influential factors include poor lighting at night, adverse weather conditions, late‐night hours (20:00–06:00), specific geographic regions (e.g., Yilan County), speed limits of 100 km/h, and vehicle types such as taxis and large trucks. The findings not only enhance the understanding of environmental influences on accident outcomes but also offer actionable insights for improving highway safety. Moreover, Taiwan’s ETC system serves as a model for countries seeking to integrate tolling infrastructure with traffic safety analytics.