Clustering nonlinearly separable and unbalanced data set

In this paper, a new clustering method, kernel based deterministic annealing (KBDA) algorithm, is developed. This development provides a possible solution for the nonlinearly separable and unbalanced data clustering problems. Basically, the kernel based method makes nonlinearly separable data set more likely linearly separable through a nonlinear data transformation from input space into a high dimensional feature space. Furthermore, the mass possibilities of different clusters are incorporated into clustering procedure, which makes KBDA capable of clustering unbalanced data set. The effectiveness of the proposed clustering method is supported by experimental results.