Premium
Hybrid dynamic learning mechanism for multivariate time series segmentation
Author(s) -
Wang Ling,
Li Kang,
Ma Qian,
Lu YanRong
Publication year - 2020
Publication title -
statistical analysis and data mining: the asa data science journal
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.381
H-Index - 33
eISSN - 1932-1872
pISSN - 1932-1864
DOI - 10.1002/sam.11448
Subject(s) - multivariate statistics , series (stratigraphy) , segmentation , computer science , cluster analysis , dynamic factor , artificial intelligence , pattern recognition (psychology) , time series , data mining , dynamic programming , machine learning , algorithm , mathematics , statistics , paleontology , biology
To improve the efficiency of segmentation methods for multivariate time series, a hybrid dynamic learning mechanism for such series' segmentation is proposed. First, an incremental clustering algorithm is used to automatically cluster variables of multivariate time series. Second, common factors are extracted from every cluster by a dynamic factor model as an ensemble description of the system. Third, this common factor series is segmented by dynamic programming. The proposed method can potentially segment multivariate time series and not only performs segmentation better on multivariate time series with a large number of variables but also improves the running accuracy and efficiency of the algorithm, especially when analyzing complex datasets.