Efficient Search of the Optimal Prefix and Suffix Invariant Size for Dynamic Time Warping

Abstract

Along with the rapid development of information technology and industrial IoT, a large amount of time series data are continuously produced in widespread fields, including medicine, finance, astronomy and so on. Accordingly, time series data mining has widely followed by researchers and time series classification (TSC) has also been attracting great interest over the past decade. Recent empirical evidence has suggested that the combination of the nearest neighbor classifiers (NNC) and the Dynamic Time Warping (DTW) is more efficiently than traditional methods. However, since time series data can not be segmented perfectly without human intervention in the raw time series, whose head and tail will tend to contribute unbalanced influence to the similarity measure, which would produce incorrect classification results. Accordingly, prefix and suffix invariant DTW (e-DTW) was proposed to provide invariance on endpoints, which can improve the classification accuracy. In this paper, we introduce RAPIDPSISSEARCH: an efficient and exact algorithm to learn the optimal prefix and suffix invariant size (OPSIS). Extensive experiments on different kinds of typical time series datasets have been conducted to demonstrate the superiority of our method.