Title of Paper:Multivariate time series segmentation approach based on hidden Markov models

Hits:

Date of Publication:2016-06-01

Journal:ICIC Express Letters

Included Journals:EI

Volume:10

Issue:6

Page Number:1427-1433

ISSN No.:1881803X

Abstract:In this paper, an approach based on hidden Markov models (HMM) is applied to segmenting multivariate time series. In this algorithm, each observation in the underlying time series is dependent on a corresponding hidden state, and the segmentation problem comes down to finding the hidden states. When related parameters are given, the states can be obtained under a maximum likelihood framework using the Viterbi algorithm. Some meaningful methods including vector autoregression (VAR) models are utilized to estimate the parameters. In this way, the state estimation step and the parameter estimation step will perform repeatedly until the convergence condition is satisfied. The segmentation procedure is evaluated by a hydrometeorological time series. © 2016 ISSN.