Predicting highway lane-changing maneuvers: A benchmark analysis of machine and ensemble learning algorithms

Understanding and predicting highway lane-change maneuvers is essential for driving modeling its automation. The development of data-based lane-changing decision-making algorithms nowadays in full expansion. We compare empirically this article different machine ensemble learning classification techniques to the MOBIL rule-based model using trajectory data European two-lane highways. analysis relies on instantaneous measurements up twenty-four spatial–temporal variables with four neighboring vehicles current adjacent lanes. Preliminary descriptive investigations by principal component logistic analyses allow identifying main intending a driver change predict two types discretionary maneuvers: overtaking (from slow fast lane) fold-down lane). prediction accuracy quantified total, lane-keeping errors associated receiver operating characteristic curves. benchmark includes model, linear discriminant, decision tree, naïve Bayes classifier, support vector machine, neural network algorithms, ten bagging stacking meta-heuristics. If provides limited accuracy, devoid bias, significant improvements. Cross validations show that selected networks from single observation both seconds advance high accuracy.