Detecting Transportation Modes Based on LightGBM Classifier from GPS Trajectory Data

Abstract

Human travel behavior can be obtained from the trajectory data generated by GPS devices, which can be reflected in different transportation modes and provide useful information for trajectory prediction, urban planning and traffic monitoring. In this article, we proposed transportation modes classification method based on Light Gradient Boosting Machine (LightGBM) to discover seven kinds of transportation modes from GPS trajectory data, including walking, cycling, taking a bus, taking a taxi, driving a car, taking the subway and taking a train. First, the original trajectories must be divided into some sub trajectories. There is only one transportation mode label in each sub trajectory. Second, the feature vector of sub trajectory is computed including eight basic and three advanced features. These basic features are distance feature, five velocity-related features and two acceleration-related features. Three advanced features are heading change rate (hcr), stop rate (sr) and velocity change rate (vcr), Final, the LightGBM classifier is used to detect the transportation modes automatically. The eXtreme Gradient Boosting (XGBoost) and decision tree are also used to verify the efficiency of our method. The experiment data are Geolife provided by Microsoft Research Asia. The results show that the LightGBM and XGBoost methods are more accurate than decision tree method and the LightGBM is better than XGBoost at the classification of car, subway and train.