Evolving Fuzzy Classification of Nonstationary Time Series
Yevgeniy V. Bodyanskiy, Ye. Gorshkov, Illya Kokshenev, Vitaliy Kolodyazhniy · 2010
The problem of adaptive segmentation of time series changing their properties at a priori unknown moments is considered. The proposed approach is based on the idea of indirect sequence clustering, which is realized with a novel robust evolving recursive fuzzy clustering algorithm that can process incoming observations online and is stable with respect to outliers that are present in real data. An application to the segmentation of a biological time series confirms the efficiency of the proposed algorithm. The approach used by the authors belongs to the class of the objective function-based algorithms that are designed to solve the clustering problem via the optimization of a certain predetermined clustering criterion, and are the best grounded from mathematical point of view. Due to its adaptive sequential form, the clustering algorithm proposed in the chapter is related to evolving fuzzy systems, and can be used for online tuning of their rule bases. Controlled Vocabulary Terms fuzzy systems; time series; workstation clusters