Application of the k-medoids partitioning algorithm for clustering of time series data

Abstract

Data clustering has been widely applied in numerous areas in order to pave the way for adequate and efficient modelling, control and operation. In the past, most of the data clustering was carried out on static data. However, wider application of time series data has increased the need for time series clustering techniques. This paper presents a comprehensive analysis of the applicability of a standard clustering algorithm, the k-medoids algorithm, for clustering of two diverse time series datasets. The k-medoids algorithm is tested on dynamic power responses of a hybrid renewable energy source plant and neuroscience spike-train data. The main stages in clustering process, that is, data processing, the selection of the optimal distance measure and the estimation of the optimal number of clusters, are analyzed in detail.