Time-series clustering by approximate prototypes

A global averaging method for dynamic time warping, with applications to clustering

“…Most clustering techniques with DTW use the K-MEDOIDS algorithm, which does not require any computation of an average [15][16][17][18]. However, K-MEDOIDS has some problems related to its use of the notion of median: K-MEDOIDS is not idempotent, which means that its results can oscillate.…”

“…Several attempts at defining an averaging method for DTW have been made, but they provide an inaccurate notion of average [14], and perturb the convergence of such clustering algorithms [15]. That is mostly why several time series clustering attempts prefer to use the K-MEDOIDS algorithm instead (see [16][17][18] for examples combining DTW and the K-MEDOIDS algorithm). Throughout this article, and without loss of generality, we use some times the example of the K-MEANS algorithm, because of its intensive use of the averaging operation, and because of its applicability to large datasets.…”

A novel two-level clustering method for time series data analysis

“…In recent years, data mining techniques and applications for time series data analysis have been gaining extensive attentions with interesting research topics on clustering, similarity search, classification and prediction, etc. (Hautamaki, Nykanen, & Franti, 2008;Kalpakis, Gada, & Puttagunta, 2001;Kaufman & Rousseeuw, 1990;Keogh, Lonardi, & Chiu, 2002;Tseng, Wang, & Lee, 2003;Vlachos, Kollios, & Gunopulos, 2002). Among these research issues, clustering analysis has been applied in a wild variety of fields such as biology, medicine, economics (Banerjee & Ghosh, 2001;Dermatas & Kokkinakis, 1996;Focardi, 2001;Guo, Jia, & Zhang, 2008;Qian, DolledFilhart, Lin, Yu, & Beyond, 2001;Tseng & Kao, 2007;Tseng & Kao, 2005), etc.…”

BEATS: Blocks of Eigenvalues Algorithm for Time series Segmentation

“…An extensive description of similarity measures can be found in [16]. DTW and CID are also used in clustering the raw time series [17] [18].…”