Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

RED CoMETS: An ensemble classifier for symbolically represented multivariate time series

Luca A. Bennett, Zahraa S. Abdallah

TL;DR

This work introduces RED CoMETS, a novel ensemble classifier for symbolically represented multivariate time series that extends the Co-eye univariate framework to multivariate data. It combines two MV extensions—Concatenating and Ensembling—with a faster univariate foundation built on SAX and SFA lenses, enhanced by a random pair selection strategy and three voting schemes. Across 111 UCR datasets, RED CoMETS demonstrates competitive accuracy with state-of-the-art baselines and notably achieves the best reported accuracy on HandMovementDirection, while offering substantial speedups over the original Co-eye. The approach enables efficient, high-accuracy MV time series classification with practical implications for domains requiring scalable analysis of complex temporal data.

Abstract

Multivariate time series classification is a rapidly growing research field with practical applications in finance, healthcare, engineering, and more. The complexity of classifying multivariate time series data arises from its high dimensionality, temporal dependencies, and varying lengths. This paper introduces a novel ensemble classifier called RED CoMETS (Random Enhanced Co-eye for Multivariate Time Series), which addresses these challenges. RED CoMETS builds upon the success of Co-eye, an ensemble classifier specifically designed for symbolically represented univariate time series, and extends its capabilities to handle multivariate data. The performance of RED CoMETS is evaluated on benchmark datasets from the UCR archive, where it demonstrates competitive accuracy when compared to state-of-the-art techniques in multivariate settings. Notably, it achieves the highest reported accuracy in the literature for the 'HandMovementDirection' dataset. Moreover, the proposed method significantly reduces computation time compared to Co-eye, making it an efficient and effective choice for multivariate time series classification.

RED CoMETS: An ensemble classifier for symbolically represented multivariate time series

TL;DR

This work introduces RED CoMETS, a novel ensemble classifier for symbolically represented multivariate time series that extends the Co-eye univariate framework to multivariate data. It combines two MV extensions—Concatenating and Ensembling—with a faster univariate foundation built on SAX and SFA lenses, enhanced by a random pair selection strategy and three voting schemes. Across 111 UCR datasets, RED CoMETS demonstrates competitive accuracy with state-of-the-art baselines and notably achieves the best reported accuracy on HandMovementDirection, while offering substantial speedups over the original Co-eye. The approach enables efficient, high-accuracy MV time series classification with practical implications for domains requiring scalable analysis of complex temporal data.

Abstract

Multivariate time series classification is a rapidly growing research field with practical applications in finance, healthcare, engineering, and more. The complexity of classifying multivariate time series data arises from its high dimensionality, temporal dependencies, and varying lengths. This paper introduces a novel ensemble classifier called RED CoMETS (Random Enhanced Co-eye for Multivariate Time Series), which addresses these challenges. RED CoMETS builds upon the success of Co-eye, an ensemble classifier specifically designed for symbolically represented univariate time series, and extends its capabilities to handle multivariate data. The performance of RED CoMETS is evaluated on benchmark datasets from the UCR archive, where it demonstrates competitive accuracy when compared to state-of-the-art techniques in multivariate settings. Notably, it achieves the highest reported accuracy in the literature for the 'HandMovementDirection' dataset. Moreover, the proposed method significantly reduces computation time compared to Co-eye, making it an efficient and effective choice for multivariate time series classification.
Paper Structure (16 sections, 1 equation, 5 figures, 4 tables, 4 algorithms)

This paper contains 16 sections, 1 equation, 5 figures, 4 tables, 4 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Univariate Foundation
  4. Pair Selection
  5. Voting
  6. Developing RED CoMETS
  7. Concatenating Approach
  8. Ensembling Approach
  9. RED CoMETS
  10. Experiments and Evaluation
  11. Experimental Design
  12. Univariate Foundation
  13. Pair Selection
  14. Voting
  15. RED CoMETS
  16. ...and 1 more sections

Figures (5)

  • Figure 1: Test accuracy critical difference diagram for random pair selection methods against Co-eye averaged over 30 resamples for each of the 85 univariate UCR datasets. Default accuracy, DTW, and ROCKET are included as benchmarks.
  • Figure 2: Pairwise comparison of total mean train and test time between Co-eye and R5% averaged over 30 stratified resamples of the 85 univariate UCR datasets.
  • Figure 3: Test accuracy critical difference diagram for proposed voting methods against Co-eye averaged over 30 resamples for each of the 85 univariate UCR datasets. Default accuracy, DTW, and ROCKET are included as benchmarks.
  • Figure 4: Test accuracy critical difference diagram for RED CoMETS variants averaged over the 23 UCR datasets.
  • Figure 5: Test accuracy critical difference diagram for RED CoMETS-3 against the state-of-the-art classifiers averaged over the 23 UCR datasets