Robust functional supervised classification for time series Articles uri icon

publication date

  • October 2014

start page

  • 325

end page

  • 350

issue

  • 3

volume

  • 31

international standard serial number (ISSN)

  • 0176-4268

electronic international standard serial number (EISSN)

  • 1432-1343

abstract

  • We propose using the integrated periodogram to classify time series. The method assigns a new time series to the group that minimizes the distance between the series integrated periodogram and the group mean of integrated periodograms. Local computation of these periodograms allows the application of this approach to nonstationary time series. Since the integrated periodograms are curves, we apply functional data depth-based techniques to make the classification robust, which is a clear advantage over other competitive procedures. The method provides small error rates for both simulated and real data. It improves existing approaches and presents good computational behavior.

keywords

  • time series; supervised classification; integrated periodogram; functional data depth; discriminant-analysis; curves