G-SteX: Greedy stem extension for free-length constrained motif discovery

Abstract

Most available motif discovery algorithms in real-valued time series find approximately recurring patterns of a known length without any prior information about their locations or shapes. In this paper, a new motif discovery algorithm is proposed that has the advantage of requiring no upper limit on the motif length. The proposed algorithm can discover multiple motifs of multiple lengths at once, and can achieve a better accuracy-speed balance compared with a recently proposed motif discovery algorithm. We then briefly report two successful applications of the proposed algorithm to gesture discovery and robot motion pattern discovery. © 2012 Springer-Verlag.