A new clustering algorithm with the convergence proof

Abstract

Conventional clustering algorithms employ a set of features; each feature participates in the clustering procedure equivalently. Recently this problem is dealt with by Locally Adaptive Clustering, LAC. However, like its traditional competitors the LAC method suffers from inefficiency in data with unbalanced clusters. In this paper a novel method is proposed which deals with the problem while it preserves LAC privilege. While LAC forces the sum of weights of the clusters to be equal, our method let them be unequal. This makes our method more flexible to conquer over falling at the local optimums. It also let the cluster centers to be more efficiently located in fitter places than its rivals. © 2011 Springer-Verlag.