Granular Sketch Based Uncertain Time Series Streams Clustering

Abstract

Uncertainty is inherent in data streams, and presents new challenges to data streams mining. For continuous arriving and huge size of data streams, it requires significantly more space to represent and cluster the uncertain time series data streams. Therefore, it is important to construct compressed representation for storing uncertain time series data. The granular sketches and buckets policy are designed through hash-compressed storage and micro clusters. Then based on the max-min cluster distance measure, an initial cluster centers selection algorithm is proposed to improve the quality of clustering uncertain data streams. Finally, the effectiveness of the proposed algorithm is illustrated through analyzing the experimental results.