Empirical mode decomposition in segmentation and clustering of slowly and fast changing non-stationary signals

Abstract

Empirical mode decomposition (EMD) is a principally new technique, intended to process various types of non-stationary signals by means of decomposing them into a set of certain functions, called “Intrinsic mode functions” (IMFs) or Empirical modes. This paper is devoted to a newly developed EMD application to Data Mining, namely, to segmentation and clustering problems. Two new algorithms of segmentation are introduced. The first one was developed for slowly changing signals and is capable of extracting monotonous segments (piecewise-polynomial segmentation) as well as other signal’s patterns. The second one, employed for fast changing signals, allows us to extract segments with different variances, energies and autoregressive model orders. Both algorithms were tested on various signals and fuzzy-clustering results of extracted segments are given. Finally, the advantages and disadvantages of these approaches are described and the possible ways of their further improvement and development are outlined.