Y.-S. Moon (Korea)
Time-series data, stream sequence matching, similar sequence matching, handheld devices, and data mining.
In recent years, real-time processing of a large amount of infinite stream data becomes a hot research issue. For handheld devices, minimization of CPU operations is the most important factor in performance. In this paper, we propose the efficient algorithms that extract sequences similar to the given query sequence from the time-series stream such as network traffic data, stock prices, and sensor data. First, we formally define the stream sequence matching that finds similar sequences from the time-series stream. Second, we propose an efficient window-based approach by using the window construction mechanism of traditional subsequence matching methods. Third, we provide the notion of a window MBR and propose two different stream sequence matching algorithms based on the notion. Fourth, we formally prove correctness of the proposed algorithms by presenting the related theorems. Last, through extensive analysis and experiments, we show that our approach improves performance significantly compared with the naive approach. Experimental results show that our window-based approach improves performance by tens to hundreds of times over the naive approach. Overall, we believe that our methods would be very suitable for handheld devices as the embedded algorithms. 503-019
Important Links:
Go Back
