关键词: 聚类, K-Modes算法, 粗糙集, 类中心, 聚类精度

Abstract: Michael K.Ng et al proposed the new K-Modes clustering algorithm. It takes the heuristic dissimilarity measure method based on the relative frequency and improves the clustering accuracy. However, when computing the attribute category frequency in each cluster, it assumes each object of the samples plays a uniform contribution to the cluster center. To consider the particular contribution of the different objects, a rough K-Modes algorithm is proposed in this paper. By a new approach based on the upper and lower approximate of rough set to measure the important level of each object in its corresponding cluster, the better clustering results can be achieved than the new K-Modes algorithm, and the computational complexity can be reduced in comparison with the improved K-Modes clustering algorithm based on rough sets of Bai Liang et al with the equivalent clustering results. The experimental results on several UCI data sets illustrate the effectiveness of the proposed algorithm.

Key words: clustering, K-Modes algorithm, rough set, cluster center, clustering accuracy