Work

Clustering Parallel Data Streams

Year: 2009
Type: book-chapter
Abstract: When our aim is to mine the similarity on the trends of data streams, correlation coefficient is a more appropriate measure for similarity between data streams than Euclidean distance used by previous data stream clustering methods. In this paper, we have developed algorithms CORREL-cluster and CORRELCOD for clustering multiple data streams based on correlation coefficients, which supports online clustering analysis over both fixed and flexible time horizons. Since data streams have high speed and massive volume, we can not retain the raw data to perform correlation analysis. We have proposed a compression scheme that supports an one-scan algorithm for computing the correlation coefficients for. We have developed an adaptive algorithm to dynamically determine the number of clusters so that CORREL-cluster can adjust to the evolving behaviors of data streams. Moreover, we have developed a novel partitioning algorithm to support clustering of arbitrary length per user's request. Experimental results on real and synthetic data sets show that our algorithms have high clustering quality, efficiency, and scalability.
Author: Yixin Chen
Language: English
Open Access status: gold
OpenAlex ID: W1540607020
FWCI: 0.433
Cites: 29
Cited by: 7
Related to: 10