Fast and accurate template averaging for time series classification

摘要: Time series data are evidently ubiquitous, as we could see them in all kinds of domains and applications. As a result, various mining tasks often performed to discover useful knowledge, including commonly like time classification clustering. Dynamic Warping (DTW) is accepted one the best available similarity measures, which has been used for distance calculation both clustering algorithms. However, its known drawback exceedingly high computational cost. Recently, condensation method through template averaging applied; each class can be represented by greatly speed up with DTW especially large datasets, trade off lower accuracies. Subsequently, attempts have made increase number representative templates boost accuracies while keeping computation complexity not too high. those algorithms still suffer from many predefined hard-to-set parameters, some require accuracy results. Therefore, this work, propose an accurate yet simple that parameter free much less time. The experiment results on 20 UCR benchmark datasets demonstrate our proposed achieve few orders magnitude speedup maintaining