Entering driver path change analysis at roundabouts using binary logit and machine learning

Abstract

Roundabouts are popular globally for their safety, capacity and environmental benefits. However, crashes remain a significant concern, especially in countries such as India. A significant number of entering vehicle drivers at roundabouts adjust their speed and take evasive action for various reasons, leading to dangerous driving behaviour. Therefore, this study has focused on analysing the driver behaviour responsible for evasive manoeuvres when vehicles enter a roundabout from the approach roads by employing a binary logit (BL) model and machine learning approaches including random forest, K-nearest neighbours and extreme gradient boosting (XGBoost). The analysis reveals that driving path distractions are highly sensitive to the vehicle gap, manoeuvring speed of entering vehicles, circulating speed of vehicles on the circulatory road, entering vehicle type, circulating traffic volume and entering traffic volume. XGBoost achieved the highest overall prediction accuracy of 75.6% among all the models evaluated. These findings enhance the understanding of driver behaviour at roundabouts and provide insights for improving safety protocols. Furthermore, evaluating the BL model and machine learning methods provides valuable guidance for selecting appropriate modelling approaches to understanding and mitigating roundabout-related crash risks.