A Flexible Clustering Algorithm Based On Statistics Bariables Granule

Cluster analysis technique, as a promising data analysis method, still faces many challenges when processing today's com- plex data set. There are mainly three effectivity problems which are considered in this paper. First, the prior knowledge, which is usually unavailable, is needed. Second, parameter tuning is diffi- cult. Third, it usually provides little knowledge about the struc- ture of clusters. Therefore, we present a clustering algorithm based on statistics granular merging of data blocks. A database is broken down into many data granule and then be gradually merged into small statistics homonomous subsets. This approach has not imposed needs for prior knowledge. Its tuning parameters is relatively easy to use. It effectively discovers the clusters and the noises in a computationally efficient way. The behavior of the proposed algorithm is illustrated on several 2D shape data sets, and the state-of-the-art performance comparing with some pop- ular algorithms is demonstrated on several real world data sets.