Investigating the Transferability of Machine Learning Methods in Short-Term Travel Time Prediction

Short-term travel time prediction is essential for Advanced Traveler Information Systems and supports proactive traffic management for road network managers. In previous studies on this topic, machine learning methods were developed for short-term travel time prediction under a wide range of conditions. However, an important practical issue that has not been adequately addressed in the literature is the application of such models across an entire network. It is rare that the extensive historical training datasets required for model training is available for all the links in the network. Transferring trained models to other links in the network is a natural way to address this issue. This paper investigates the transferability of different machine learning methods in short-term traffic prediction using travel time data collected from a real-world network. The result of the experiments shows that it is possible to transfer machine learning models trained on a link to the other links under certain conditions based on comparing the similarity of observable factors of the training and target links; however, further research is needed to explore in more details the factors that affect transferability.

• Supplemental Notes:
  • This paper was sponsored by TRB committee ABJ30 Standing Committee on Urban Transportation Data and Information Systems.
• Authors:
  • Luan, Jianlin
  • Guo, Fangce
  • Polak, John
  • Hoose, Neil
  • Krishnan, Rajesh
• Conference:
  • Transportation Research Board 97th Annual Meeting
  • Location: Washington DC, United States
  • Date: 2018-1-7 to 2018-1-11
• Date: 2018

Language

• English

Media Info

• Media Type: Digital/other
• Features: Figures; Maps; References; Tables;
• Pagination: 13p

Subject/Index Terms

• TRT Terms: Machine learning; Mathematical prediction; Network links; Travel time
• Subject Areas: Data and Information Technology; Transportation (General);

Filing Info

• Accession Number: 01658357
• Record Type: Publication
• Report/Paper Numbers: 18-02742
• Files: TRIS, TRB, ATRI
• Created Date: Jan 29 2018 10:27AM