Abstract
The analysis and estimation of lane change (LC) behavior are essential for autonomous vehicles (AVs) to predict other vehicles' intentions and avoid accidents. Since the LC intention is easily affected by various features, the feature selection and LC modeling greatly influence the prediction accuracy and interpretability. Therefore, a binary logistic regression LC model with a mean impact value (MIV) method to select features is proposed for accurate prediction. First, the related features are classified as individual, microscopic, and macroscopic levels. Then they are ranked and analyzed by the MIV method. Next, the closely related features are selected and used as input to the logistic regression model for LC intention prediction. As a result, a highly interpretable LC model is built with a prediction performance of around 80%. This paper benefits the quantification and explanation of the influences of different levels' features on LC intention and lays a solid foundation for the AVs to predict the LC behavior.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call