Accurate subsequence matching on data stream under time warping distance

With ever more increases in both storage and computational power, mining data in real time has become more and more practical and rapidly consumed a large fraction of research works on data streams. Particularly for time series data, subsequence matching under dynamic time warping (DTW) distance has proven to work exceptionally well, but with higher time and space complexities, thus posing a much challenging problem to work on streaming data. Recent work has proposed a solution to this problem with only linear time and space. However, we demonstrate that it may give inaccurate results, and then propose a novel accurate subsequence matching (ASM) algorithm that eliminates this discrepancy. We further demonstrate utilities of our work on a comprehensive set of experiments that guarantees an improvement in accuracy while maintaining the same time and space complexities.