Shape-based representation and abstraction of Time Series Data along with a Dynamic Time Shape Wrapping as a dissimilarity measure

Fatma-Ezzahra Gmati, Salem Chakhar, Wided Lejouad Chaari, Mark Xu

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

This paper proposes a Time Series Shape (TSS) based framework for time series representation and abstraction. The paper also introduces a Dynamic Time Shape Wrapping (DTSW), which is a shape extension of the well-known Dynamic Time Wrapping (DTW) dissimilarity measure. By jointly supporting representation and abstraction, TSS and its related dissimilarity measure DTSW can be applied in hybrid time series data mining tasks, especially those involving both rule induction and classification. The paper also compares the capabilities of TSS and Piecewise Linear Approximation (PAA) representation in a classification task. Results show that TSS has the same dimensionality reduction power as PAA. This means that TSS is able to maintain the same classification accuracy as PAA, with an additional time series abstraction capability. The results also indicate that DTSW is able to successfully quantify the comparison between TSS abstractions.
Original languageEnglish
Title of host publicationProceedings of the 26th IEEE International Conference on Automation and Computing (ICAC'21), Portsmouth, UK, 2-4 September, 2021
PublisherInstitute of Electrical and Electronics Engineers
Publication statusAccepted for publication - 6 Jun 2021
Event26th IEEE International Conference on Automation and Computing (ICAC'21) - Portsmouth, United Kingdom
Duration: 2 Sep 20214 Sep 2021

Conference

Conference26th IEEE International Conference on Automation and Computing (ICAC'21)
CountryUnited Kingdom
CityPortsmouth
Period2/09/214/09/21

Keywords

  • Time Series Representation
  • Time Series Abstraction
  • Dissimilarity Measure
  • Dynamic Time Shape Wrapping
  • zeroembargo

Fingerprint

Dive into the research topics of 'Shape-based representation and abstraction of Time Series Data along with a Dynamic Time Shape Wrapping as a dissimilarity measure'. Together they form a unique fingerprint.

Cite this