成对约束的属性加权半监督模糊核聚类算法  

Semi-supervised mercer-kernel based fuzzy clustering algorithm with pairwise constraints and attribute weighted

作　　者：贺杨成[1] 王士同[1] 江南[1] 

机构地区：[1]江南大学数字媒体学院,江苏无锡214122

出　　处：《计算机工程与应用》2011年第24期136-138,162,共4页Computer Engineering and Applications

基　　金：国家自然科学基金(No.60704047);国家自然科学基金重大研究计划(No.9082002);国家高技术研究发展计划(863)(No.2007AA1Z158)~~

摘　　要：在机器学习和数据挖掘中,带约束的半监督聚类是一个活跃的研究领域。为了利用约束条件获得表现更优异的聚类效果,提出了一种成对约束的属性加权半监督聚类算法,该方法充分考虑了属性间的不平衡性,在传统模糊聚类算法中融合半监督学习机制并通过Mercer核把原始的观察空间映射到高维特征空间。实验结果表明,该算法优于相似的成对约束的竞争群算法(PCCA)。Clustering with constraints is an active area in machine learning and data mining.A semi-supervised mercer-kernel based fuzzy clustering algorithm with pairwise constraints and attributes weighted is proposed which incorporates both semi-supervised learning technique and the kernel method into the traditional fuzzy clustering algorithm.The proposed algorithm performs clustering in high feature space mapped by a mercer kernels and considers the imbalance between the attributes ful-ly.This algorithm experimentally outperforms a similar semi-supervised fuzzy clustering approach,i.e Pairwise Constrained Competitive Agglomeration（PCCA）.

关 键 词：半监督聚类 成对约束 核 模糊聚类 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]