Prediction of Expressway Sections Vulnerable to Icing-Related Traffic Accidents Considering Road Characteristics and Climate Conditions in the Republic of Korea

Abstract In this study, the risk level of icing-related traffic accidents for each Korean expressway section is evaluated using geometric characteristics, traffic volume, and local climatic conditions to efficiently allocate the limited safety management resources according to the level of risk. To this end, data on road icing accidents that had occurred on the entire expressways in Korea over the last 10 years were collected. The expressways were divided into approximately 180,000 sections. Meteorological, traffic, and geometric data were collected for each section to construct a GIS-based database. Multicollinearity analysis and data preprocessing using EasyEnsemble were conducted. Consequently, a Random Forest model was trained to estimate feature importance and the risk level of road icing accidents for each road section. The model identified average temperature, the difference between average temperature and dew point, winter precipitation days, traffic volume per lane, longitudinal slope, and radius of curvature as the most influential variables in that order. Sections with higher risk levels of icing-related accidents were predominantly concentrated in the western regions of Korea, where the difference between average temperature and dew point, precipitation, and traffic per lane volume were relatively high. Additionally, it was confirmed that the road sections identified by the proposed model as vulnerable to icing accidents accounted for approximately 2.1 times more actual icing-related accidents than those designated by the Korean government. The feature importance values and corresponding risk levels of icing accidents for each road section predicted in this study are expected to be used for future prevention and mitigation of winter road traffic accidents.