Comparison of Interpolation Methods for Missing Traffic Volume Data

This paper compares eleven algorithms for imputing missing traffic volume data recorded by automatic traffic detectors; eight of these algorithms are described in previous literature, and three are novel. These algorithms are compared using loop detector data from the Dallas, Texas region, by artificially deleting a sample of data, calibrating the eleven algorithms using the remaining observations, and imputing the deleted sample. Three types of data loss were simulated: random losses, continuous loss of data for an extended period of time, and systematic loss of data at the same times each day. The algorithms were then compared by their fidelity in reconstructing the ``missing'' data. Of the eleven algorithms, those based in linear regression were found to be most accurate; however, their requirements are highest in terms of input data, which should be considered for practical implementation. By contrast, those based on historical data can be applied more often, but are significantly less accurate. Significant differences were also seen between the different data loss scenarios, with much higher bias for systematically missing data.

Language

  • English

Media Info

  • Media Type: DVD
  • Features: Figures; References; Tables;
  • Pagination: 16p
  • Monograph Title: TRB 90th Annual Meeting Compendium of Papers DVD

Subject/Index Terms

Filing Info

  • Accession Number: 01334660
  • Record Type: Publication
  • Report/Paper Numbers: 11-3757
  • Files: TRIS, TRB
  • Created Date: Mar 31 2011 8:11AM