Comparing time series segmentation methods for the analysis of transportation patterns with smart card data

Time series clustering is important in the analysis of action. In the domain of transportation, it is especially important. It allows an understanding of people's activities within a time period. In this article, a method is presented for the segmentation of time series that takes advantage of cross-correlation distance, hierarchical clustering, and considers the separation of positive/negative correlations in order to understand temporal patterns of users. This method consists of two steps: 1. Combining cross correlation distance and hierarchical clustering to obtain cluster groups, and 2. dividing these groups into smaller sized groups by separating cross correlation parameters, "correlation coefficient" and "lag". Considering that dynamic time warping is a common method to measure time series distance, the clustering results are compared between dynamic time warping distance and cross correlation distance. After a small pedagogical example, we develop a program by R to validate the method on a real data set. The results of the real data demonstrate that this method precisely segments the time series. This comparison result also demonstrates the advantage of using cross correlation distance in the domain of public transportation.


  • English

Media Info

  • Pagination: 19p

Subject/Index Terms

Filing Info

  • Accession Number: 01643835
  • Record Type: Publication
  • Source Agency: ARRB
  • Report/Paper Numbers: CIRRELT-2017-28
  • Files: ATRI
  • Created Date: Aug 21 2017 9:29AM