English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Conventional time series classification approaches based on bags of patternsor shapelets face significant challenges in dealing with a vast amount offeature candidates from high-dimensional multivariate data. In contrast, deepneural networks can learn low-dimensional features efficiently, and inparticular, Convolutional Neural Networks (CNN) have shown promising results inclassifying Multivariate Time Series (MTS) data. A key factor in the success ofdeep neural networks is this astonishing expressive power. However, this powercomes at the cost of complex, black-boxed models, conflicting with the goals ofbuilding reliable and human-understandable models. An essential criterion inunderstanding such predictive deep models involves quantifying the contributionof time-varying input variables to the classification. Hence, in this work, weintroduce a new framework for interpreting multivariate time series data byextracting and clustering the input representative patterns that highlyactivate CNN neurons. This way, we identify each signal's role anddependencies, considering all possible combinations of signals in the MTSinput. Then, we construct a graph that captures the temporal relationshipbetween the extracted patterns for each layer. An effective graph mergingstrategy finds the connection of each node to the previous layer's nodes.Finally, a graph embedding algorithm generates new representations of thecreated interpretable time-series features. To evaluate the performance of ourproposed framework, we run extensive experiments on eight datasets of theUCR/UEA archive, along with HAR and PAM datasets. The experiments indicate thebenefit of our time-aware graph-based representation in MTS classificationwhile enriching them with more interpretability.

MTS2Graph: Interpretable Multivariate Time Series Classification with Temporal Evolving Graphs