Journal of Korean Society of Transportation ISSN:1229-1366(Print) 2234-4217(Online) 대한교통학회지

Recently the number of car accidents has been increasing when the precipitation such as rain or snow has occurred. To reduce the number of car accidents, not only continuous activity for road maintenance but also driver’s perception of road risk is necessary. In this paper, we engineered spatio-temporal features using weather and spatial information collected in the area of Seoul, Gyeonggi-do and Incheon, and trained random forest model predicting road surface conditions up to accuracy of 95%. Spatial index, which is a feature of explaining spatial information, is more efficient in proportion to the size of trained data, and achieved at most 5% of accuracy increase rate compared to a prediction model where spatial index has not been trained. Because the prediction model predicts not point but area, our research contributes to providing road surface condition information to any drivers in the trained area. Finally, expanding the collecting area of road surface conditions in order to train other surrounding environments, the prediction model is expected to predict road surface conditions of wider area.