关键词: 计算机应用, C均值算法, 初值敏感, 邻域密度

Abstract: C-means Clustering Method (CM) is a widely for data clustering, which is sensitive to the initial cluster centers and easily leads to local optimum. To solve this problem, an Optimal Initialization-based C-means Method (OI-CM) is proposed. First for each point in the dataset, the neighborhood and neighborhood density are calculated, and the point with the maximum neighborhood density is selected as the first cluster center. Then, the point with the maximum neighborhood density from the rest datasets is selected as the next cluster center, whose neighborhood must have little coupling degree with the neighborhoods of existing cluster centers. This procedure is continued until all the cluster centers are selected. Finally, the CM is utilized to cluster the datasets with the selected cluster centers. Experimental results on simulated and UCI datasets show that the proposed OI-CM can effectively solve the sensitivity defect of the traditional CM to initial cluster centers, and has superior performance than other three global CMs.

Key words: computer application, C-means method, initial value sensitivity, neighborhood density