Skip to content

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

License

Notifications You must be signed in to change notification settings

zy18811/RED-CoMETS

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

30 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

RED-CoMETS

RED CoMETS: An Ensemble Classifier for Symbolically Represented Multivariate Time Series

AALTD 2023 | arXiv:2307.13679 (preprint)

Multivariate time series classification is a rapidly growing area of research that has numerous practical applications in diverse fields such as finance, healthcare, and engineering. The problem of classifying multivariate time series data is challenging due to the high dimensionality, temporal dependencies, and varying lengths of the time series. This paper proposes a novel ensemble classifier called RED CoMETS. The proposed method builds upon the success of Co-eye, an ensemble classifier for symbolically represented univariate time series, and extends it to handle multivariate data. RED CoMETS is evaluated on benchmark datasets from the UCR archive and achieves competitive accuracy with state-of-the-art techniques in multivariate settings and yields the highest accuracy reported in the literature for the 'HandMovementDirection' dataset. The proposed method also offers a significant reduction in computation time compared to classic Co-eye, making it an efficient and effective option for multivariate time series classification.

Please cite as:

@inproceedings{10.1007/978-3-031-49896-1_6,
  author    = {Bennett, Luca A. and Abdallah, Zahraa S.},
  title     = {RED CoMETS: An Ensemble Classifier for Symbolically Represented Multivariate Time Series},
  booktitle = {Advanced Analytics and Learning on Temporal Data},
  publisher = {Springer Nature Switzerland},
  address   = {Cham},
  year      = {2023},
  pages     = {76-91}
}

Results

Test accuracy is given for all experiments.
Timing results have only been recorded for Co-eye, R5%, R10%, R15%, and R20%.

Univariate Foundation

Co-eye

Random Pair Selection

Voting Methods

RED CoMETS

Code

An implementation of RED CoMETS is also available on aeon.

Requirements

Loading Datasets

UCR time series classification archive datasets in .arff format.

from Code.Utilities.utils import DatasetLoader

dataset = '/path/to/dataset/folder'

# Univariate Datasets
X_train, y_train, X_test, y_test = DatasetLoader(dataset).load_uv_dataset_to_numpy()

# Multivariate Datasets
X_train, y_train, X_test, y_test = DatasetLoader(dataset).load_mv_dataset_to_numpy()

Co-eye

from Code.Univariate_Foundation.coeye import coeye

[...] # load dataset

"""
Args:
    p_length (float): percentage of timeseries length used to determine number of SAX and SFA lenses
    voting_method  (string):  {'uniform', 'meanmax', 'validation'}
"""
accuracy = coeye(X_train, y_train, X_test, y_test)

Univariate Foundation

from Code.Univariate_Foundation.univariate_foundation import univariate_foundation

[...] # load dataset

"""
Args:
    p_length (float): percentage of timeseries length used to determine number of SAX and SFA lenses
    voting_method  (string):  {'uniform', 'meanmax', 'validation'}
"""
accuracy = univariate_foundation(X_train, y_train, X_test, y_test, p_length, voting_method)

RED CoMETS

Name Approach Sub-Approach Voting Method 1 Voting Method 2
RED CoMETS-1 Concatenating n/a Uniform n/a
RED CoMETS-2 Concatenating n/a Mean-Max n/a
RED CoMETS-3 Concatenating n/a Validation n/a
RED CoMETS-4 Ensembling 1 Uniform n/a
RED CoMETS-5 Ensembling 1 Mean-Max n/a
RED CoMETS-6 Ensembling 2 Uniform Uniform
RED CoMETS-7 Ensembling 2 Uniform Mean-Max
RED CoMETS-8 Ensembling 2 Mean-Max Mean-Max
RED CoMETS-9 Ensembling 2 Mean-Max Uniform
from Code.RED_CoMETS.red_comets import red_comets

[...] # load dataset

"""
Args:
  id_number (int): RED CoMETS-<ID>
"""
accuracy = red_comets(X_train, y_train, X_test, y_test, id_number)

Reproducing the Experiments

We used 85 univariate and 26 multivariate datasets from the UCR timeseries classification archive.

Co-eye and Univariate Foundation

univariate_experiments.py

RED CoMETS

multivariate_experiments.py

Acknowledgements

We thank all the people who have contributed to and who maintain the UCR time series classification archive. Critical difference diagrams in our paper showing were produced using code from Ismail Fawaz et al. (2019). Our implementation of the SAX transform uses code modified from saxpy.

About

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

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages