Abstract
Road traffic accidents (RTAs) are increasingly becoming a global scourge, leading to numerous mortalities and morbidities. The global statistics on RTAs-induced mortalities are worrisome, as RTAs are among the top eight causes of death globally. While there is increasing research interest in applying machine learning (ML) and deep learning (DL) algorithms to predict and model RTAs, there is a dearth of studies that review and organize the existing literature to identify the efficacy and performance of such models, the algorithms, features and datasets used, and the challenges associated with using ML and DL for modelling and predicting RTAs. Thus, this study adopted the Preferred Reporting Items for Systematic Reviews and Meta-Analysis to determine factors associated with RTAs and identify RTAs predictive models, their performance, weaknesses and strengths. The study shows that human factors, weather conditions, road conditions, the day of the week, cognitive impairment of road users, travelling hours, traffic flow and events are among the important factors associated with road traffic accidents. The findings revealed that some ML and DL algorithms used for RTAs modelling and prediction include logistic regression, CatBoost, Support Vector Machines, k-nearest neighbour, long short-term memory, generative adversarial networks, gated recurrent unit and convolutional neural networks. Understanding the impact of these factors and RTAs predictive models may assist policymakers, transportation safety designers, researchers, traffic agents and law enforcement agencies in developing interventions and preventive measures to reduce RTAs while improving road safety.