Travel Mode Detection using Smartphone GPS Data: a Comparison between Random Forest and Wide-and-Deep Learning

Artificial intelligence methods are widely used in travel mode detection based on passively collected data (GPS trajectory data). Various travel modes have been analyzed in the literature, but the underground metro mode was never investigated due to poor cellular reception underground. This study, to the best knowledge of the authors, is the first in the U.S. to infer underground metro modes using land use information. The authors innovatively line-up the multimodal transportation network to the GPS trajectories to infer the closeness to the nearby rail (both underground and aboveground) and bus lines. This paper compares two mode detection models: random forests and wide-and-deep learning. Compared to random forests, the wide-deep model combines the advantages of both “wide” and “deep” models to be able to make sufficient generalizations using multi-layer deep learning and capture the exceptions using the wide single-layer model. The model is empirically tested on a GPS dataset collected in the Washington D.C. and Baltimore metropolitan regions. The empirical test showcases the superior goodness-of-fit of the proposed wide-deep learning model.

  • Supplemental Notes:
    • This paper was sponsored by TRB committee ABJ30 Standing Committee on Urban Transportation Data and Information Systems.
  • Corporate Authors:

    Transportation Research Board

  • Authors:
    • Yang, Di
    • Xiong, Chenfeng
    • Tang, Liang
    • Zhang, Lei
  • Conference:
  • Date: 2019


  • English

Media Info

  • Media Type: Digital/other
  • Features: Figures; References;
  • Pagination: 5p

Subject/Index Terms

Filing Info

  • Accession Number: 01697516
  • Record Type: Publication
  • Report/Paper Numbers: 19-00444
  • Files: TRIS, TRB, ATRI
  • Created Date: Dec 7 2018 9:29AM