Efficient stream subsequence matching algorithms for handheld devices on streaming time-series data

초록

In recent years, the real-time processing of infinite online stream data is becoming a hot research issue. For the handheld devices dealing with the stream data, minimizing repetitive CPU operations such as multiplications is a major factor for their performances. In this paper, we propose efficient algorithms for finding similar subsequences from streaming time-series data such as stock prices, network traffic data, exchange rates, and sensor network data. First, we formally define the problem of similar subsequence matching from streaming time-series data, which is called the strewn subsequence matching in this paper. Second, based on the window construction mechanism adopted by the previous subsequence matching algorithms in the non-streaming environment, we present an efficient window-based approach that minimizes CPU operations required for stream subsequence matching. Third, to make sure the window-based approach to be practical, we propose a notion of window MBR and present two stream subsequence matching algorithms based on the notion. Fourth, we formally prove correctness of the proposed algorithms. Finally, through a series of analyses and experiments, we show that our algorithms significantly outperform the naive algorithm. To our knowledge, there has been no algorithm to efficiently handle the stream subsequence matching problem except the naive one. Analysis results show that the numbers of multiplications of our algorithms are reduced by the factor of several hundreds than the naive one. Experiment results of measuring the elapsed execution times show that the performances of our algorithms are improved by up to several tens of times over the naive one. We believe that our window-based algorithms are excellent choices for embedded stream subsequence matching in handheld devices.

키워드