一种基于成员选择的簇加权聚类集成算法  

A cluster-weighted clustering ensemble algorithm based on member selection

作　　者：徐森[1] 高婷 徐秀芳[1] 许贺洋 郭乃瑄 卞学胜 花小朋[1] 陈致远 XU Sen;GAO Ting;XU Xiu-fang;XU He-yang;GUO Nai-xuan;BIAN Xue-sheng;HUA Xiao-peng;CHEN Zhi-yuan(Faculty of Information Engineering,Yancheng Institute of Technology,Yancheng 224002,China;Key Laboratory of Computer Network and Information Integration(Southeast University),Ministry of Education,Nanjing 211189,China)

机构地区：[1]盐城工学院信息工程学院,江苏盐城224002 [2]计算机网络和信息集成教育部重点实验室(东南大学),南京211189

出　　处：《控制与决策》2024年第12期4136-4140,共5页Control and Decision

基　　金：国家自然科学基金项目(62076215,62301473);中央高校基本科研业务费专项资金项目(K93-9-2022-03);江苏省高等教育厅面上项目(23KJB520039);江苏省网络与信息安全重点实验室项目(BM2003201);江苏高校“青蓝工程”项目;盐城市基础研究计划项目(YCBK2023008);盐城工学院研究生培养创新工程项目(SJCX23-XY060)。

摘　　要：聚类集成算法在数据挖掘和模式识别等领域应用广泛.现有的聚类集成算法虽取得了显著的进展,但鲜有同时考虑如何处理冗余成员和关注成员内部多样性的算法.对此,设计一种簇的不确定性度量指标,并提出一种基于成员选择的簇加权聚类集成算法.首先,利用平均差异性度量和筛选聚类成员,并引入信息熵衡量簇的不确定性,给簇赋予相应的权重;其次,在基于成员选择的簇加权共协矩阵和高置信度矩阵的基础上构建增强矩阵;最后,在增强矩阵上执行层次聚类算法得到最终的聚类集成结果.采用多个UCI数据集进行实验,将所提算法与主流的聚类集成算法进行比较,实验结果表明,所提出的算法可以获得更好的聚类集成效果,且具有较高的鲁棒性和稳定性.Clustering ensemble algorithms are widely used in fields such as data mining and pattern recognition.Although the existing clustering ensemble algorithms have made significant progress,few algorithms consider how to deal with redundant members and pay attention to the diversity within members at the same time.In this paper,we design an uncertainty metric for clusters,and propose a cluster-weighted clustering ensemble algorithm based on member selection.Firstly,the average difference is used to measure and screen the cluster members,and the uncertainty of the cluster is measured by information entropy,and the corresponding weight is given to the cluster.Then,the enhanced matrix is constructed on the basis of the cluster-weighted co-association matrix and the high-confidence matrix based on member selection.Finally,the hierarchical clustering algorithm is executed on the enhancement matrix to obtain the final clustering ensemble result.Experiments are carried out on multiple UCI datasets,and the proposed algorithm is compared with the mainstream clustering ensemble algorithms,and the experimental results show that the proposed algorithm can obtain better clustering integration effect and has high robustness and stability.

关 键 词：聚类集成 数据挖掘 成员选择 簇加权 信息熵 共协矩阵 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]