Roundabouts are widely embraced for their perceived safety advantages over other types of unsignalized intersections. However, there has been an observed increase in crash rates at roundabouts over time in Jordan. This paper delves into modeling traffic crash severity at roundabouts, considering various factors such as weather, lighting, vehicle characteristics, geometric features, and driver age and gender. To comprehensively analyze roundabout crashes in Jordan, we constructed rule-based classifiers and Random Forest models after balancing the dataset. Rule-based models offer interpretability, albeit with some simplicity trade-off, while Random Forest models provide deeper analysis but require additional explanation. Presenting both outputs to subject matter experts and policymakers facilitates a holistic understanding of factors contributing to roundabout crashes in Jordan. Subsequently, CN2 results revealed that injury severity crashes are influenced by the time of the day, driver age, day of the week, speed, and number of vehicles involved. On the other hand, property damage-only crashes are affected by the number of lanes, time of the day, type of driver fault, lighting conditions, speed, and day of the week. The RF model analysis unveiled crucial factors influencing crash severity in roundabouts, notably the varying impact of driver age, time of day, the number of vehicles involved, seasonality, and vehicle speed. This proposed approach is promising, comprehensive, and not only enhances the understanding of roundabout crashes but also informs the development of effective, localized safety interventions.
Read full abstract