ROAD TRAFFIC ACCIDENTS PREDICTION USING MACHINE LEARNING METHODS

Abstract

Road traffic accidents are identified as a significant societal issue based on extensive and comprehensive research in public health and traffic safety. Such incidents lead to significant negative outcomes, including human casualties, economic consequences, vehicle damage, and significant medical costs. Developing predictive models is crucial for identifying risk factors associated with accidents, thereby improving understanding of accident causes and enabling more effective prevention interventions. The occurrence of traffic accidents is influenced by a multitude of factors, including driver behavior, vehicle characteristics, weather conditions, road volume, road geometry, type of road, road conditions, speed limits, frequency of police controls, etc. In this paper, machine learning (ML) techniques are used to develop the traffic accident prediction model due to the non-linear relationship between input and output variables. The research investigates the influence of certain input variables on the number of traffic accidents, as their optimal choice significantly affects the prediction performance. The random forest, support vector machine, and neural networks are employed for data preprocessing and model development. Statistical indicators are used to evaluate the performance of the developed models. Based on the obtained performances, the developed ML models accurately predict the number of traffic accidents.