Abstract: On the basis of improvements on defects in data stream clustering algorithm based on density grid, a data stream clustering algorithm was proposed which improved D-Stream algorithm. The algorithm set density threshold of grid cell dynamically by statistics on density of grid cell and number of clusters. To increase the precision of cluster boundary, a non-uniform division was employed on the grid boundary cell. The result of experiments on synthetic and real data set shows that the algorithm has fast processing speed and the ability to detect dynamic changes of data for data stream clustering, and improves clustering quality.

Key words: Data mining, Data stream, Cluster analysis, Density grid, Non-uniform division