Machine Learning-Based Intention Recognition for Right-Turning Vehicles at Signalized Intersections

Signalized intersections serve as critical hubs in urban road networks. At intersections without dedicated right-turn phases, frequent interactions occur between right-turning vehicles and pedestrians/non-motorized vehicles. Suboptimal interactions may lead to traffic conflicts, significantly compromising travel safety and operational efficiency. This study categorizes right-turning vehicle intentions into three types: full-stop yielding, deceleration yielding, and non-yielding behaviors. Influencing factors are classified into agent-related factors and environmental factors, with input features for intention recognition models being selected through filter methods. Three intention recognition models-Support Vector Machine, Random Forest, and Logistic Regression-are developed to identify right-turning vehicle intentions. Through comprehensive evaluation metrics including accuracy and precision, comparative analysis reveals that the Logistic Regression model demonstrates optimal overall performance in precisely capturing right-turning vehicle intentions.