标签分布熵正则的模糊C均值平衡聚类方法

Label distribution entropy regularized fuzzy C-means algorithm for balanced clustering

许多应用场景要求每个类别的数量相对平衡,而传统模糊C均值(FCM)聚类算法无法实现此功能.为此,利用标签信息构造标签分布熵评价聚类的平衡度,然后将标签分布熵、模糊隶属度矩阵与标签矩阵之间的平方损失同时引入到传统FCM中,进而提出一种标签分布熵正则的模糊C均值平衡聚类方法(FCMLDE).同时,利用迭代方法和增广拉格朗日乘数法设计该模型的优化算法.最后,利用6个真实数据集进行聚类实验,结果表明,所提方法在聚类性能和平衡性能上均具有很好的优势.

Many application scenarios require that the number of each category is relatively balanced,and the traditional fuzzy C-means(FCM)clustering algorithm cannot achieve this function.For this reason,we first design a label distribution entropy by using the label information,which can evaluate the balance degree of clustering.Then,the label distribution entropy and the square loss between the fuzzy membership matrix and the label matrix are simultaneously introduced into the traditional FCM,and then a fuzzy C-means balanced clustering method based on the regular label distribution entropy(FCMLDE)is proposed.Besides,this paper designs an optimization algorithm to solve the proposed model through the iterative strategy and the augmented Lagrange multipliers method.Finally,clustering experiments are performed using six real data sets,and the results show that the proposed method has good advantages in clustering performance and balance performance.