An efficient implementation of anytime k-medoids clustering for time series under dynamic time warping

Abstract

Time series clustering is one of the crucial tasks in time series data mining. So far, time series clustering has been most used with Euclidean distance. Dynamic Time Warping (DTW) distance measure has increasingly been used as a similarity measurement for various data mining tasks in place of traditional Euclidean distance due to its superiority in sequence-alignment flexibility. However, there exist some difficulties in clustering time series with DTW distance, for example, the problem of speeding up DTW distance calculation in the context of clustering. So far, there have been two proposed methods for time series clustering with DTW and both of them work in batch scheme. Recently, Zhu et al. proposed a framework of anytime clustering for time series with DTW which uses a data-adaptive approximation to DTW. In this paper, we present an efficient implementation of anytime K-medoids clustering for time series data with DTW distance. In our method, we exploit the anytime clustering framework with DTW proposed by Zhu et al., apply a method for medoid initialization, and develop a multithreading technique to speed-up DTW distance calculation. Experimental results on benchmark datasets validate our proposed implementation method for anytime K-medoids clustering for time series with DTW.