Data Stream (Time series database) has been widely used in aspect of Web monitor and sensor net. Mining on data streams has been a brand new research field in recent years, as the representation of data streams turned to be the critical problem of this field. In this paper, we propose a novel segmentation algorithm, which can represented data stream in an effective and compact fashion. We first introduce the general view of Piecewise Linear Representation which represents data streams, typical segment algorithms on data streams and their features. After that, we present some ideas on improving performance and compress efficiency of segmentation algorithm using PLR, and introduce a fast Sliding Window based algorithm on data stream segmentation using linear regression. Detailed experiments have been conducted to evaluate the performance of proposed method, and the results show the superiority of our algorithm.
赵哲、孙婷、陈立军、崔斌
展开 >
北京大学计算机科学技术系 北京 100871
Data stream Segment algorithm Linear approximation Online Compress Threshold
NETL
