ABSTRACT
In the literature different distance measures and clustering models have been proposed for classifying observed or transformed time series (observation-based-clustering). This chapter shows some distance measures for observed or transformed time series; It presents some models for clustering time series able to capture the instantaneous and longitudinal characteristics of the observed time series. The chapter reports some applicative examples. A “direct” approach for clustering time series can be based on a comparison of the observed time series or a suitable transformation of the observed time series. The distance measures 1d(), 2d(), 3d(), 4d() are useful for identifying similar geometric profiles of the time series and in general, clustering based on these distances will be mainly dominated by local comparisons. In the literature there are various distances for time series based on dynamic time warping (DTWDTW). This is a technique used to find an optimal alignment between two given time series across time points under certain restrictions.
