Short-Term Freeway Traffic Prediction Using Gradient Boosting Regression Trees

Short-term traffic predictive performance is considerably influenced by the neighboring traffic condition. The prediction accuracy has been enhanced by adding the upstream and downstream traffic information to the advanced prediction models. In this study, the gradient boosting decision trees GBDT), a relatively new and robust ensemble learning method, is proposed to perform the freeway short-term traffic prediction, which can identify the importance of variables on the predictor responses and improve the prediction accuracy. Based on the boosting technique, GBDT algorithm adds new simple decision trees sequentially, trained with the error of the previous whole ensemble model at each iteration, which can capture the complex traffic nonlinearity and produce an outstanding predictive performance. Moreover, the interaction between the input variables and models can be well interpreted with the relative importance of variables. The parameters of GBDT algorithm are optimized to produce higher accuracy with fewer iterations, and prevent over-fitting. The GBDT prediction models with different combinations of input variables referring to the neighboring traffic information are performed to explore the influences of upstream and downstream traffic condition on the traffic prediction of the current site. Experiment results show that the predictive performance can be enhanced by adding the neighboring traffic information to the models. GBDT models perform better than the classical SVM models in the short-term traffic prediction of freeway. The relative importance of variables varies considerably among different models, and the importance of upstream traffic condition is not equal to that of downstream.

• Supplemental Notes:
  • This paper was sponsored by TRB committee ABJ70 Standing Committee on Artificial Intelligence and Advanced Computing Applications.
• Corporate Authors:

  Transportation Research Board

  500 Fifth Street, NW
  Washington, DC  United States  20001
• Authors:
  • Yang, Senyan
  • Wu, Jianping
  • Du, Yiman
  • He, Yingqi
  • Chen, Xu
• Conference:
  • Transportation Research Board 96th Annual Meeting
  • Location: Washington DC, United States
  • Date: 2017-1-8 to 2017-1-12
• Date: 2017

Language

• English

Media Info

• Media Type: Digital/other
• Features: Figures; References; Tables;
• Pagination: 15p
• Monograph Title: TRB 96th Annual Meeting Compendium of Papers

Subject/Index Terms

• TRT Terms: Algorithms; Decision trees; Machine learning; Programming (Mathematics); Regression analysis; Traffic forecasting; Traffic models
• Subject Areas: Data and Information Technology; Highways; Planning and Forecasting;

Filing Info

• Accession Number: 01627786
• Record Type: Publication
• Report/Paper Numbers: 17-03249
• Files: TRIS, TRB, ATRI
• Created Date: Feb 27 2017 5:12PM