Similarity search for time series based on efficient warping measure

Abstract

Similarity search is one of the most important tasks in time series data mining, and similarity measure between time series is a basic work. Dynamic time warping (DTW) is often used to compute distance between two time series by warping time axes to match the same shapes. However, its high computation complexity is an obstacle to similarity search based on DTW. To address the issue of similarity search using DTW in time series data mining, an efficient dynamic time warping based on backward strategy and search scope reduction to find the optimal warping path is proposed. At the same time, a small threshold value in the efficient time warping is used to stop similarity measure in advance and to fast expel the dissimilar time series. In this way, a novel similarity search method for time series based on the efficient dynamic time warping without manual intervention is formed. The proposed method searches similar time series with more accuracy and improves its computation speed by the two search scope reductions. The results of experiments on time series datasets demonstrate that in contrast to classical dynamic time warping, the new method can be used to search the similarity in time series databases fast and accurately.