Prediction of the Road Accidents Severity Level: Case of Saint-Petersburg and Leningrad Oblast

Abstract

This article examines the factors influencing the severity of road accidents in St. Petersburg and Leningrad oblast for 2015–2023. The study is carried out on the analysis of 69190 road accidents and 6 groups of factors using the logit model and testing the oversampling technique to predict the probability of severe injuries and fatal cases after road accidents. The main factors in the study were lighting, deficiencies in road maintenance, and mean of transport. In particular, the logit model made for a joint sample on Saint – Petersburg and Leningrad oblast showed that the absence of lighting increases the probability of a serious accident by 19.6%, the presence of a vehicle such as a truck or motorcycle in a traffic accident increases the probability by 10.9%, and the presence of fog raises the probability by 17.6%. The usage of Synthetic Minority Over-sampling Technique (SMOTE) did not lead to a significant increase in the prediction accuracy of the models. The results of the study can be useful for organizing safe traffic in the city and providing recommendations for road users and public officials involved in improving the city’s infrastructure.

Keywords

• logit model
• machine learning
• road safety
• smote
• traffic accident