Abstract: The fixed weights are adopted in the traditional rough K-means algorithm to represent the different approximations of the clusters, but it is always difficult to predefine the optimal weights with little priori knowledge before clustering. Therefore,an improved rough K-means algorithm based on self-adaptive weights was proposed in this paper.The new method computes the weights for every data according to the current clustering state and no more does rely on the initial weights. Furthermore, the self-adaptive weights arc obtained from the Gaussian distance ration in cluster approximation, which can lead to the more accurate clustering results. The experiments indicate that the rough K-means based on self-adaptive weights is an effective rough clustering algorithm.

Key words: Clustering,Rough sets,Rough K-means,Self-adaptive weight