A Density Grid-Based Clustering Algorithm for Uncertain Data Streams

Abstract

This paper proposes a grid-based clustering algorithm Clu-US which is competent to find clusters of non-convex shapes on uncertain data stream. Clu-US maps the uncertain data tuples to the grid space which could store and update the summary information of stream. The uncertainty of data is taken into account for calculating the probability center of a grid. Then, the distance between the probability centers of two adjacent grids is adopted for measuring whether they are "close enough" in grids merging process. Furthermore, a dynamic outlier deletion mechanism is developed to improve clustering performance. The experimental results show that Clu-US outperforms other algorithms in terms of clustering quality and speed.