Modeling tactical lane-change behavior for automated vehicles: A supervised machine learning approach

In recent years, due to growing interest in automated driving, the need for better understanding the humans driving behavior, and particularly the lane changing and car following behavior, has further increased. Despite its great importance, lane changing has not been studied as extensively as longitudinal behavior and remains one of the most challenging driving behavior maneuvers to understand and to predict. Drivers take into account many factors while making a tactical decision, which cannot be precisely represented by the conventional rule-based models. In this paper, we compare the results of different supervised machine learning classifiers to better understand the lane change decision of drivers using the NGSIM database. For this aim, after choosing the relevant features, the ones which contribute the most to the model were chosen with the help of feature importance analysis. Afterward, the training dataset was used to train the model with naive Bayes, support vector machines, logic regression, nearest neighborhoods, decision trees, extra trees and random forest classifiers. The accuracy of predictions for test dataset indicates that extra trees classifier, decision trees and random forest had the best performance in predicting the lane change decisions of human drivers.