Clusters of multivariate stationary time series by differential evolution and autoregressive distance

Abstract

Clustering MTS is a difficult task that has to be performed in several application fields. We propose a method based on the coefficients of vector autoregressive (VAR) models and differential evolution (DE) that may be applied to sets of stationary MTS. Results from a simulation experiment that includes both linear and non linear MTS are displayed for comparison with genetic algorithms (GAs), principal component analysis (PCA) and the k-means algorithm. Part of the Australian Sign Language (Auslan) data are examined to show the comparative performance of our procedure on a real world data set. © 2011 Springer-Verlag Berlin Heidelberg.