基于样本加权的可能性模糊聚类算法  

A Sample-Weighted Possibilistic Fuzzy Clustering Algorithm

作　　者：刘兵[1] 夏士雄[1] 周勇[1] 韩旭东[1] 

机构地区：[1]中国矿业大学计算机科学与技术学院,江苏徐州221116

出　　处：《电子学报》2012年第2期371-375,共5页Acta Electronica Sinica

基　　金：国家自然科学基金(No.50674086);国家博士后科学基金(No.20070421041)

摘　　要：可能性模糊聚类算法解决了噪音敏感和一致性聚类问题,但算法假定每个待分析样本对聚类的贡献相同,导致离群点或噪声点对算法的干扰较强,算法迭代次数过大.为此,提出一种基于样本加权的可能性模糊聚类算法,新算法具有更快的收敛速度,对标准数据集和人工数据集加噪后的测试结果表明,该算法具有更强的鲁棒性,在有效降低时间复杂度的同时能够取得较好的聚类准确率.The possibilistic fuzzy clustering algorithm overcomes the problem of sensitivity to noises and coincident clusters, but it assumes the contribution of each sample is equal, which leads to strong impact from outliers or noises and too many iterations. For this reason,this paper proposes a novel faster possibilistic fuzzy clustering algorithm based on the sample-weighted idea. The re- sults of the experiments on standard data sets and synthetic data sets show that the sample-weighted algorithm is more robust against noises and outliers and reduces the time complexity effectively, and can obtain good clustering accuracy at the same time.

关 键 词：样本加权 可能性C-均值聚类 可能性模糊聚类 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]