Spatial Road Safety Analysis, Machine Learning for Accident Prediction, and Safe Routing in Accident- and Congestion-Prone Road Networks: The Case of Addis Ababa City, Ethiopia

Abstract

Road accidents are a significant global issue leading to economic setbacks, fatalities, injuries, and traffic congestion. The growth of urban areas and the increasing number of vehicles impact transportation safety, with low-income countries experiencing much higher death rates compared to high-income countries. Predicting and averting road traffic accidents (RTAs) and congestions are crucial, particularly in developing countries. This study responds to the global challenge of road accidents and traffic congestion, concentrating on low-income countries such as Ethiopia, with a specific focus on Addis Ababa, where the fatality rates are disproportionately high. The research reviews global literatures concerning road accidents and congestion, and assesses prediction techniques aimed at improving road safety and reducing traffic congestion. It utilizes GIS-based spatial statistics, crash rate analysis, advanced machine learning (ML) algorithms, and Network-Based Analysis of Optimization Algorithms. The objectives are to identify existing challenges and suggest ways to prevent road network accidents and congestion. Spatial crash rate analysis and Getis-Ord-Gi* statistics, combined with spatial machine learning (ML) using a random forest model (RF), are employed in this research to increase prediction accuracy. The dataset used includes crash records from 2014 to 2019 in Addis Ababa comprising 64, 878 crash incidents, road networks, and boundary shapefiles. Spatial network analysis is used to determine safe routes, avoiding accident and congestion-prone road segments. The key findings of this study include the accident hotspots identification through a combined crash rates analysis and the successful prediction of RTAs, along with determining of safe routes in Addis Ababa. The research maps crash locations and identifies significant RTA segments associating them with the underlying road network. This analysis reveals notable hotspots on 10 roundabouts/squares, 33 road segments, and 3 intersections, with the gravest accident hotspot averaging 37.5 annual number of crashes per kilometer during the 2014-2019 periods. The study employs a spatial ML RF model, achieving 78% predictive accuracy, while spatial network analysis determines alternative safe routes. These results demonstrate the methods strong capability contributing to improved traffic safety by alerting drivers and travelers to potential accident prone road segments and high traffic congestion risks. vi Results from the first method showed that hotspots in Addis Ababa are primarily located at intersections and roundabouts in densely populated sub-cities of the downtown area, where frequent conflicts occur between motor vehicles and other road users. The second method’s outcomes highlight the need to determine safe routes based on crash frequencies and the likelihood of accidents and congestion. Therefore, both the crash rate and predicted likelihood of RTAs served to determine the safest travel route. This study underscores the importance of integrating spatial analysis techniques, optimization algorithms, and ML methods to reduce traffic congestion and promote road safety. Spatial and ML predictive models are crucial in identifying accident-prone areas, while network analysis provides recommendations to safer routes. The findings demonstrate that these methods enable transportation authorities and stakeholders to quickly identify unsafe locations within the road network, effectively prioritize hotspot areas, evaluate the association between traffic accidents and congestions, and provide the safer route options to drivers. This method contributes research on road safety and aids efforts in transport planning. This research benefits transportation authorities, drivers, and travelers by offering practical solutions for improving road safety, particularly in limited-resource urban areas. By prioritizing safety measures and increasing awareness of accident-prone locations, the study aims to reduce traffic accidents and alleviate congestion. It recommends further refining spatial and machine learning predictive models with additional variables and real-time data sources expanding the research to other urban areas to assess scalability, and exploring emerging technologies for data collection and analysis to improve road safety and traffic management.