Multivariate time series classification based on M-histograms and multi-view learning

Abstract : Univariate time series (UTS) classification has been reported in several papers, where various efficient models have been proposed. Such models are often inadequate for multivariate time series (MTS) classification. MTS emerged with the multiplication of sensors that record large amounts of high-dimensional data, characterized by several dimensions , variable lengths, noise and correlations between dimensions. MTS classification is a challenging task, and few works have been devoted to complex data. In this paper, we propose a novel subspace model that combines M-histograms and multi-view learning together with an ensemble learning technique to handle MTS classification task. The M-histograms is a statistical tool, efficient for data visualization, that can reveal mutual information from different dimensions. Thus it can be suitable for MTS data encoding. Multi-view learning is known as data integration from multiple feature sets, as it is the case with MTS data, and multiple views also provide complementary information. The new combined model provides a MTS encoding that can outperform other time series encoding such as Symbolic Aggregate approXimation (SAX) w.r.t. to the experimental comparison that was conducted. We also benchmark our method with some state-of-art methods devoted to MTS, and discuss the results obtained and the main properties of our model.
Document type :
Preprints, Working Papers, ...
Complete list of metadatas

Cited literature [24 references]  Display  Hide  Download

https://hal.archives-ouvertes.fr/hal-02162067
Contributor : Angéline Plaud <>
Submitted on : Friday, June 21, 2019 - 1:37:23 PM
Last modification on : Thursday, June 27, 2019 - 1:34:49 AM

File

plaud_etal_ecml_2019.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-02162067, version 1

Citation

Angéline Plaud, Engelbert Nguifo, Jacques Charreyron. Multivariate time series classification based on M-histograms and multi-view learning. 2019. ⟨hal-02162067⟩

Share

Metrics

Record views

18

Files downloads

51