Identifying Transportation Modes Using Gradient Boosting Decision Tree

Abstract

Identifying the transportation modes could be applicable to many applications including personalized recommendation, transportation planning. The existing studies had not fully considered the impact of geographical information. In this paper, we propose a novel approach to detect transportation modes from massive trajectories using Gradient Boosting Decision Tree (GBDT), which adopted and estimated the impact of geographical information to achieve a better performance. In the experiments, we conduct the performance evaluation using the Geolife dataset which collected by 182 users over five years. The dataset contains 8347 trajectories with transportation mode such as driving, taking a bus, riding a bike and walking. 60% of trajectories are randomly chosen as training dataset, and then we tested on the remaining dataset. The experimental results showed that our proposed approach considering geographical information by using gradient boosting decision tree method achieve the precision of 84%, with the maximum increase of 6.83% to the traditional identifying transportation modes method. In addition, the geographical information contributed over 12% to improve the precision of recognition.