Transportation modes recognition using a Light Gradient Boosting Machine

Abstract

To investigate different traffic modes for resident’s travel trajectories, a classification model was constructed based on Light Gradient Boosting Machine (LightGBM) to categorize transportation modes according to resident’s GPS trajectories. First, basic trajectory features were extracted, and then more features were obtained using geographic information of public transit network (i.e., Fréchet distance). Subsequently, the features were normalized and screened by the decision tree model. Finally, the screened features were trained and predicted by the model, and a stable prediction result was attained with a five-fold cross-validation method. Results show that geographic information of public transit network could optimize the model’s prediction accuracy. The proposed GPS trajectory recognition method achieved an accuracy of about 90%, which is superior to other machine learning classification models.

Publication
Journal of Harbin Institute of Technology, vol. 51(9) pp. 96-102, September 2019.