Authors: Mirlei Moura Da Silva; Rodrigo Fernandes De Mello; Ricardo Araújo Rios
Addresses: Department of Computer Science, Federal University of Bahia, Salvador, BA, Brazil ' Institute of Mathematics and Computer Science, University of São Paulo, São Carlos, SP, Brazil ' Department of Computer Science, Federal University of Bahia, Salvador, BA, Brazil
Abstract: Time series clustering aims at designing methods to extract patterns from temporal data in order to organise series according to their similarities. In general, most of researches either perform a preprocessing step to convert time series into attribute-value matrices to be later analysed by traditional clustering methods, or apply measures specifically designed to compute the similarity among time series. We noticed two main issues in such studies: 1) clustering methods do not take into account stochastic and deterministic influences inherent in real-world time series; 2) similarity measures tend to look for recurrent patterns, which may not be available in stochastic time series. In order to overcome such drawbacks, we present a new clustering approach that considers both influences and a new similarity measure to deal with purely stochastic time series. Experiments provided outstanding results, emphasising time series are better clustered when their stochastic and deterministic influences are properly analysed.
Keywords: time series; clustering; similarity measure; time series decomposition; stochastic component; feterministic component.
International Journal of Computational Science and Engineering, 2020 Vol.21 No.3, pp.394 - 417
Received: 16 Aug 2018
Accepted: 08 Nov 2018
Published online: 13 Mar 2020 *