关键词: 模糊C均值算法, 模糊聚类, 核函数, 有效性指标, 聚类个数估计

Abstract: Using kernelized metric of compactness and separation, this paper proposed a new clustering validity index named KKW, and obtained the optimized cluster number. Besides, the KKW index was used in the modified kernel fuzzy clustering (MKFCM) algorithm. As mapped by modified Mercer kernel functions, the data set shows new features never showed before. MKFCM algorithm was applied to the data set Wine and glass. For every clustered class, MKFCM has overall accuracy higher than 90%;as to the incomplete data set Wisconsin Breast Cancer, difference is 4.72%. The modified kernel clustering algorithm is faster than the classical algorithm in convergence and more accurate in clustering. The results of simulation experiments show the feasibility and effectiveness of the modified kernel clustering algorithm.

Key words: Fuzzy C-Mean (FCM) algorithm, fuzzy clustering, kernel function, validity index, clusters number estimation