A clustering algorithm for data stream based on grid-tree and similarity

Algorithms based on k-means are incompetent to find clusters of arbitrary shapes, and the number of clusters needs to be pre-specified. Moreover, most grid-based clustering algorithms can not deal with boundary points accurately. To address these issues, a novel approach based on density gird-tree and similarity, DGTSstream, is proposed. In DGTSstream, each new data record will be mapped into the gird-tree, and sporadic grids will be removed through setting update cycle and noise density threshold. The average density is exploited to design density threshold. This algorithm repeatedly seeks a maximum density grid without cluster flag, which will be used as a starting point for finding clusters according to depth-first strategy. Finally, the similarity is adopted to deal with the boundary points. Experimental results show that our algorithm can find clusters of arbitrary shapes, and has better clustering accuracy and efficiency.