Keito KATAOKA

Evaluation of Traffic Accidents and Stuck Vehicle Occurrence Risks on Expressways Under Snowfall Conditions

Kazushi SANO

This study aimed to clarify the mechanisms behind winter highway traffic accidents and vehicle stacking using probe data from connected cars, and to quantitatively assess the risk of these rare events. First, probe data were integrated with meteorological, traffic, road alignment, and spatial data to organize the occurrence patterns of both events. The results showed that both were associated with reduced speed and unstable driving behavior, but differed in their dominant factors and occurrence processes.
Next, binary logistic regression models were developed to predict traffic accidents and vehicle stacking as rare events. To address class imbalance, multiple synthetic data generation methods were introduced. The results confirmed that meteorological conditions such as snowfall, driving conditions such as average speed, traffic composition including the proportion of large vehicles, road geometry, and spatial factors all contributed to both events. In addition, the choice of synthetic data generation method and decision threshold strongly affected the balance between suppressing false negatives and false positives, highlighting the importance of interpreting occurrence probabilities as risk indicators and adjusting decision conditions according to operational objectives.
A discrete choice model was then constructed to analyze traffic accidents and vehicle stacking within a common framework. The results indicated that although both events can arise from shared hazardous conditions, differences in factors such as snowfall intensity, longitudinal gradient, and heavy vehicle volume lead to different outcomes. This model also enabled a relative assessment of which event was more likely under conditions where multiple hazardous events developed simultaneously.
Finally, occurrence probabilities from each model were defined as risk indicators and tested against actual cases. When a probability of 0.5 or higher was treated as high risk, many cases showed that continuous or increasing snowfall raised risk by worsening road surface conditions and destabilizing traffic flow. This supported the validity of the proposed risk indicator.