Machine learning for predictions of road traffic accidents and spatial network analysis for safe routing on accident and congestion-prone road networks
Yetay Berhanu, Dietrich Schröder, Bikila Teklu Wodajo, Esayas Alemayehu
Abstract
Road traffic accidents (RTAs) and the resulting traffic congestion are global concerns mainly in metropolitan environments. The need for road safety is directly correlated with the rapidly increasing impact of urbanization on infrastructure and day-to-day living. In this study, we have introduced an innovative approach integrating a Random Forest (RF) model, crash rates, and spatial network analysis to provide safe route recommendations for drivers aiming to reduce RTAs and congestion. Based on historical accident data from 2014–2019, the analysis of the RF model and crash rate served as a prediction of the likelihood and occurrence of RTAs. In applying the spatial network analysis, lower predicted crash counts from spatial joining were taken into consideration, as well as areas with lower crash rates that have had fewer incidents in the past. An alternative safe route and an optimum route that covers 32.27 km in 50.78 mins of travel time and 28.6 km in 41.58 mins of travel time were successfully identified, respectively. Having demonstrated 78 % predictive capability on the target variable, the RF model has proved its worth. Analyzing historically lower accident counts on segments leading to minimal crash rates validates the accuracy of the identified safe route. This advanced method significantly aids in improving traffic safety by making drivers and travelers aware of potentially high rates of accidents and traffic congestion on road segments. It assists travelers with their trip planning to anticipate potential risks and suggest safer alternate routes, making it a valuable contribution to the field. • This study combined spatial crash rate, RF ML, and network analyses to predict accidents, and suggest safer routes. • Using road characteristics and historical accident data to pinpoint high-risk areas, the RF model demonstrated strong predictive capability. • Identified safe route and optimal routes avoiding accident-prone and congested areas. • Refine the model by validating in diverse environments, updating data, integrating real-time information, and improving crash data recordings.