Using Visualization to Evaluate the Performance of Algorithms for Multivariate Time Series Classification

In this paper, we use visualization tools to give insight into the performance of six classifiers on multivariate time series data. Five of these classifiers are deep learning models, while the Rocket classifier represents a non-deep learning approach. Our comparison is conducted across fifteen data...

Full description

Saved in:
Bibliographic Details
Main Authors: Edgar Acuña, Roxana Aparicio
Format: Article
Language:English
Published: MDPI AG 2025-04-01
Series:Data
Subjects:
Online Access:https://www.mdpi.com/2306-5729/10/5/58
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:In this paper, we use visualization tools to give insight into the performance of six classifiers on multivariate time series data. Five of these classifiers are deep learning models, while the Rocket classifier represents a non-deep learning approach. Our comparison is conducted across fifteen datasets from the UEA repository. Additionally, we apply data engineering techniques to each dataset, allowing us to assess classifier performance concerning the available features and channels within the time series. The results of our experiments indicate that the ROCKET classifier consistently achieves strong performance across most datasets, while the Transformer model underperforms, likely due to the limited number of instances per class in certain datasets.
ISSN:2306-5729