Investigation of the Effect of Slope and Road Surface Conditions on Traffic Accidents Occurring in Winter Months: Spatial and Machine Learning Approaches

Abstract

Winter weather can cause extremely dangerous road conditions. In order to analyze traffic accidents occurring in winter months in more detail, it is very important to evaluate the slope and the condition of road surfaces together. For this purpose, this study analyzed the accidents that occur during these months in Erzurum, one of the cities in Turkey with long winter months. A total of nine different classes of road conditions were created according to these two factors. In accordance with these classes, the accidents were analyzed using machine learning algorithms, and the success of the classification was analyzed. As a result of the analysis, it was found that the J48 algorithm gave more accurate results. J48 processes both continuous and categorical attributes and is a decision tree algorithm that can effectively manage missing data. According to the results of this algorithm, a map of accident density in the city was created using ArcGIS 10.5 software. Accordingly, it was found that the highest risk of accidents during the winter months occurred on road sections with a slope of more than 6% and covered with ice. Another important result of the study is that the slope of the road is a more effective factor than the surface condition.