Open AccessClustering multivariate time series based on Riemannian manifold
Author(s) -
Sun Jiancheng
Publication year - 2016
Publication title -
electronics letters
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.375
H-Index - 146
ISSN - 1350-911X
DOI - 10.1049/el.2016.0701
Subject(s) - cluster analysis , mathematics , geodesic , statistical manifold , covariance , series (stratigraphy) , pattern recognition (psychology) , manifold (fluid mechanics) , riemannian manifold , covariance matrix , k medians clustering , distance matrices in phylogeny , tangent space , correlation clustering , algorithm , artificial intelligence , cure data clustering algorithm , computer science , statistics , mathematical analysis , information geometry , combinatorics , geometry , mechanical engineering , paleontology , scalar curvature , curvature , biology , engineering
An approach for clustering multivariate time series (MTS) is presented in cases of variable length, noisy data or mix of different type variables. First the covariance matrices are estimated which is used as a feature to represent the MTS, then project the covariance matrices from a Riemannian manifold into a tangent space and finally carry out the clustering based on a distance matrix. In this procedure, a geodesic‐based distance is also introduced for measuring the similarity between the MTS samples. The proposed approach on a chaotic MTS with known clustering structure, namely Lorenz system is evaluated.