Development of a model for risk of occurrence of stuck on winter highways

Abstract

In this study, we adopted binary logistic regression analysis to develop a stuck occurrence risk detection model that considers road surface conditions and defined the calculated probability of vehicle stuck occurrence as the Stuck Occurrence Risk Index. Since vehicle stuck events are rare, the data involved is imbalanced, and we examined methods to handle such imbalanced data. The road surface conditions were estimated based on observed vehicle speeds and meteorological information. Through binary logistic regression analysis, we developed a model that calculates the Stuck Occurrence Risk Index using information from driving support systems, such as tire slip control and anti-skid control, obtained from connected cars, as well as vehicle operating conditions, traffic and weather information, road geometry, and road surface conditions, and clarified the relationships between these factors.