Using motif information to improve anytime time series classification
Nguyen Quoc Viet, Duong Tuan Anh · 2013
Anytime algorithm for time series classification requires the ordering heuristic of the instances in the training set. To establish the ordering, the algorithm must compute the distance between every pair of time series in the training set. And this step incurs a high computational cost, especially when Dynamic Time Warping distance is used. In this paper, we present an method to speed up the computation of this step. Our method hinges on the ordering of time series motifs detected by a previous task rather than ordering the original time series. Experimental results show that our new ordering method improves remarkably the efficiency of the anytime algorithm for time series classification without sacrificing its accuracy.