Analysis of Traffic Conflicts at Roundabout Entrances and Exits – A Machine Learning Approach for Enhanced Safety

As a component of the urban road network, roundabouts play a crucial role in ensuring operational efficiency. The safety performance of roundabouts significantly impacts overall traffic safety, making it necessary to conduct safety analysis and evaluation. This study utilises UAV to capture video of vehicle trajectory at roundabouts, employing the time to collision (TTC) index and vehicle evasive actions to identify and analyse traffic conflicts. A real-time traffic safety evaluation method has been developed using machine learning algorithms, including random forest (RF), support vector machine (SVM), extreme gradient boosting (XGBoost) and decision tree (DT) model. This method aims to analyse the relationship between traffic states and conflicts, providing insights into potential safety risks in various traffic conditions. The four machine learning algorithms trained a total of 12 models, with RF demonstrating superior training effectiveness. It achieved high accuracy in predicting traffic conflict areas at the entrances and exits of a roundabout, with a prediction accuracy of 0.86 and an AUC (area under the receiver operating characteristic curve) of 0.88. In addition, this paper further explores the relationship between traffic conflict and state. The results show that traffic flow, speed, density, speed standard deviation and vehicle type ratio have a significant relationship to traffic conflict. This research provides valuable insights for transportation authorities to understand the nature of traffic conflicts at roundabouts, enabling them to implement appropriate early warning systems and management strategies.