Development and Study of a Combined Algorithm for Temporal Series Clustering

Abstract

This paper offers a combined algorithm for temporal series clustering based on the primary clustering of the points of each temporal series, and then on the secondary clustering of a set of temporal series by a set of parameters that are statistical characteristics of the primary clusters. The paper describes the experiments to reveal the minimum set of parameters for the temporal series clustering, and compares the results of the algorithm operation for various methods of clustering. The proposed algorithm allows the clustering of temporal series with various numbers of points, with various time scales; it unites into one cluster the temporal series with a consistent similarity of parts of the graphs with the accuracy up to their contraction, extension, shift along the OX and OY axes, and Symmetries.