Road traffic accidents (RTAs) claim approximately 1.3 million lives worldwide annually. In response to this critical issue, the United Nations (UN) has set a target under the Sustainable Development Goals (SDGs) to halve the number of fatalities and injuries from RTAs by 2030 under SDG No 3 Target 3.6. RTAs have significant social and economic impacts on individuals, families, and societies, with the most severe consequence being the loss of human lives. The risk factors affecting RTA fatalities are varied and dynamic, generally categorized into human, road, vehicle, and environmental factors. Understanding the evolution of these factors is crucial for developing effective preventive and reactive measures. Machine learning techniques have become increasingly popular for identifying risk factors and predicting RTA fatalities due to their flexibility and fewer assumptions compared to traditional statistical models. This study employs comprehensive UK road safety datasets spanning 1992 to 2021, utilizing eight machine learning algorithms to develop classification models. Through stratified 5-fold cross-validation and Synthetic Minority Oversampling Technique (SMOTE), the research analyzed accident and vehicle records totaling over 5.4 million instances. Random Forest models demonstrated superior performance, achieving an Area Under Curve (AUC) between 81.28% and 82.46% and accuracy ranging from 96.65% to 97.25%. The results indicate that Random Forest models outperformed other algorithms in prediction capability. The top five identified risk factors were: urban or rural areas of the road, the number of vehicles involved, speed limit, month of the accident, and the vehicle's maneuver before the accident. Notably, these risk factors remained relatively stable over the past 30 years. This information can help relevant authorities, such as transport and police agencies, to design more targeted and up-to-date interventions to reduce the number and fatality rate of RTAs, potentially contributing to the UN's SDG target of halving road traffic injuries and fatalities by 2030.
Lee et al. (Wed,) studied this question.