A0828
Title: Finding time series motif by using swarm intelligence method
Authors: Hendri Sutrisno - Academia Sinica (Taiwan) [presenting]
Frederick Kin Hing Phoa - Academia Sinica (Taiwan)
Abstract: Time series motif discovery has been one of the most discussed problem domains in data mining. Most of the methods proposed for discovering time series motifs are computationally exhaustive, mainly on more extensive time-series data. We propose a swarm intelligence method to approximate the motif location. In the methodology, the solutions in the search space were clustered into several sub-optimum groups based on an automatic clustering mechanism to enable the local-global search strategies. The local search mechanism bounds the search space based on the clustering result into regions to improve exploitation ability, and the global search mechanism promotes information changing between the local best solutions to improve the exploration ability. The experiment results on both synthetic and real datasets reconfirm that our method can speed up dramatically compared to the current techniques for discovering the time series motifs.
