半监督平衡化模糊C-means聚类

传统模糊C-means聚类(FCM,fuzzy C-means)在处理非平衡数据集时,由于相异类中所含样本数量差异较大,导致类间权值不平衡和"均匀效应",从而易产生聚类错误.另外,FCM属于无监督方法,无法更好地利用已知的部分类标记信息引导聚类.为解决这两方面问题,提出一种半监督的平衡化模糊C-means聚类(SBFCM,semi-supervised balanced fuzzy C-means)方法.SBFCM在FCM目标函数的基础上加入了对聚类模糊隶属度矩阵的近似正交约束和半监督约束,从而得到了新的聚类目标函数.实验结果表明,相比于FCM,SBFCM能有效缓解由"均匀效应"导致的聚类错误现象,并能有效地利用部分先验类标记信息,从而可获得更好的聚类效果.

A semi-supervised balanced fuzzy C-means clustering method

When the traditional fuzzy C-means clustering(FCM, fuzzy C-means) deals with unbalanced data sets, clustering error has often occurred since the large difference in the number of samples contained in different classes, such that the weights between the classes are unbalanced and "uniform effects". In addition, FCM is an unsupervised method that does not effectively utilize known partial class labels to guide clustering. In order to solve the two problems, this paper proposes a semi-supervised balanced C-means clustering(SBFCM, semi-supervised balanced fuzzy C-means) method. Based on the FCM objective function, SBFCM adds orthogonal and semi-supervised constraints to the fuzzy membership matrix, and thus obtains a new clustering objective function. The experimental results show that compared with FCM, SBFCM can effectively alleviate the clustering error caused by "uniform effects", and can better utilize some prior class labels so that better clustering effects can be obtained.