Evaluation of Machine Learning Models for Discretionary Lane Changing Decisions

This study presents the evaluation of selected machine learning models in predicting discretionary lane changing decisions in German and California highways. Fuzzy inference system, support vector machine and artificial neural network models were analyzed. First, these models were trained, and their decisions evaluated with the observed driver's behavior in the highD vehicle trajectory data collected in German highways. Then, the predictions made by two fuzzy inference system models with the highD data set from Germany and the NGSIM data set from California were compared. The results indicated that none of the models tested were superior than the others, and these models may need to incorporate one or more input variables to make better lane changing decisions in different driving conditions.