Comparison of Machine Learning Algorithms for Predicting Lane Changing Intent

The ability to predict the intent of drivers from surrounding vehicles to change lanes is key to risk assessment and early danger warning systems. Since lane change trajectories are highly nonlinear, many studies have been performed on various machine learning algorithms using different features to predict a driver's intent to change lanes. However, these algorithms use various features that cannot be obtained from the ego vehicle's view point. In this paper, we define features that can be detected from the ego vehicle via on-board sensors and vehicle-to-vehicle communication (V2V). Gini Impurity is used to select the most appropriate features. Additionally, we compare several machine learning algorithms, including random forest (RF), support vector machine (SVM), long short term memory (LSTM), and gated recurrent unit (GRU), to find the best algorithm to predict lane changes. We evaluate the performance of these four algorithms on the Next Generation Simulation (NGSIM) dataset, which was collected by the Federal Highway Administration of the U.S. Department of Transportation. We use the I-80 dataset to train a lane changing prediction model and the US-101 dataset to test it. The test results indicate that RF had the best accuracy of the tested algorithms with an accuracy of 82 % in predicting lane changing intent.