改进的半监督模糊聚类算法.pdfVIP

第25 卷 第1 期 2010 年1 月 Vol. 25 No. 1 Cont rol a nd Decisi on Jan . 20 10 : 100 12 0920(2010) 012 01152 06 , , ( 江南大学 信息工程学院, 江苏 无锡 214 122) : 针对G i a 等近期提出的利用点对约束的半监督模糊聚类算法, 其约束项与竞争聚类算法( CA) 的 标函数 之间数量级不一致, 造成隶属度调整过度的问题, 在重新定义 标函数的基础上提出一种改进算法, 约束惩罚函数采 用约束点对中两个样本新的联合表达式, 使数量级与经典模糊聚类算法一致. 实验结果显示, 新算法的约束项与CA 标函数之间能很好地协调合作, 并能通过对模糊隶属度的适度调整, 实现更准确的聚类. : 半监督模糊聚类; 竞争聚类算法; 点对约束; 惩罚代价函数 : TP3 11 : A An impr oved semi2sup er vised fuzzy clustering algorithm GAO Cui2f an g , WU Xi ao2j un , ZH AN G S on g2shun ( School of Info mation Enginee ing, Jiangnan Unive sity, Wuxi 214122, China. Co espondent : WU Xiao2jun , E2mail : wu_xiaoj un@yahoo. com. cn) Ab stra ct : T he sem2i sup e vised fuzzy clu ste ing with pai wise con st aints ecently p oposed by G i a is analyzed. The disag eement on the magnitude o de betw een const aint te m and objective function of comp etitive cluste ing algo ithm( CA) is the main cau se fo t he ove adju st ment of memb e ship values. Aiming at this p oblem, an imp oved algo ithm is p oposed based on a edefined obj ective function . Its penalty cost function int oduces a new co2 exp ession of t wo samples in the p ai s, which has the same magnit ude o de as that of the typical fuzzy clu ste ing . Expe iment al esults show that the con st aint te m of the new algo ithm can achieve good ag eement and coope at ion with the obj ect ive function of CA , and can p oduce mo e accu ate cluste ing esults by mode ately enhancing o educing the ambiguous memb e ship s. Key wor ds: Sem2i supe vised fuzzy clu ste ing ; CA algo ithm; P ai wise con st aint