Adjustable Probability Density Grid-Based Clustering for Uncertain Data Streams

Abstract

Most existing traditional grid-based clustering algorithms for uncertain data streams that used the fixed meshing method have the disadvantage of low clustering accuracy. In view of above deficiencies, this paper proposes a novel algorithm APDG-CUStream, Adjustable Probability Density Grid-based Clustering for Uncertain Data Streams, which adopts the online component and offline component. In online component, the Probability Density Grid Clustering Feature is defined to store the summary information of uncertain data streams, and the time decay factor that introduced into the definition of the probability can reduce the influence of outdated data on clustering results. Init_clustering algorithm is called at special time interval in offline component, it first adjusts sparse probability density grid unit and updates the clustering feature of all probability density grid units. For dense probability density grid, we find and merge all dense or medium neighboring probability density grids connected with this dense probability density grid, and then the Init_clustering results is obtained. Finally APDG-CUStream returns final clustering results. The experimental results show that APDG-CUStream algorithm can accurately and rapidly obtain the clustering results with arbitrary shapes and also get better clustering quality.