Efficient Online Subsequence Searching in Data Streams under Dynamic Time Warping Distance

Data streams of real numbers are generated naturally in many applications. The technology of online subsequence searching in data streams becomes more and more important for monitoring and mining stream data. Due to its capability of handling temporal distortions in sequences, dynamic time warping (DTW) distance is a widely used similarity measure for time-series pattern matching. Unfortunately, because of the high computational complexity of DTW, no one has proposed efficient methods for online subsequence searching under DTW distance, especially over high speed data streams. In this paper, we observe that some important properties of DTW can be used to eliminate a lot of redundant computations. Based on these properties, an efficient batch filtering method for online subsequence searching in data streams is proposed. The experimental results show that when no global path constraint is used, the proposed method outperforms the best known method up to 25 times in terms of throughput. When global path constraint is considered, the proposed method can still outperform the rival method under most of the settings of the global path constraint, although our method does not exploit any information about the constraint.