A Gravity-Base Objects´ Weight Clustering Algorithm

Although many clustering algorithms have been proposed so far, seldom was focused on weight of objects. They totally or partially ignore the fact that not all data objects are equally important with respect to the clustering purpose, and that data objects which are close and dense should have more influence to sub-cluster centroid. we think that the similarity or dissimilarity of two objects is not depend on all attributes with special need, some attributes should be use to measure the dissimilarity, others attributes should impact the centroid in other way. A new weighted clustering algorithm call GBWCA is proposed to deal with different objects weight. To evaluate the proposed algorithm, we use some real and artificial dataset to compare with other algorithm, we present performance comparisons of GBWCA versus k-means and show that GBWCA is consistently superior.