K-近邻估计协同系数的协同模糊C均值算法

Novel collaboration fuzzy C-means algorithm with K-nearest neighbor method determined Collaboration Coefficient

针对现有协同模糊C均值算法(CFC)的协同系数不能充分描述数据子集间协同关系的问题,提出K-近邻估计协同系数的协同模糊C均值算法(β_K-CFC)。用模糊C均值算法(FCM)求出各数据子集的隶属度和聚类中心;其次设定近邻数,求出子集在各聚类中心处的密度,形成密度矩阵;根据密度矩阵的相关性设定变化的协同系数;最后用变化的协同系数进行协同聚类。实验证明K-近邻估计协同系数的协同模糊C均值算法(β_K-CFC)能够充分描述数据子集间的协同关系,聚类性能较好。

The collaboration coefficient of Collaboration Fuzzy C-Means(CFC)algorithm is always determined by prioriknowledge and remains constant during collaboration stages, with an inadequate using of the collaborative relationship. Inorder to circumvent this limitation, a novel CFC algorithm with K-nearest neighbor method determined collaboration coefficientis developed. Firstly, fuzzy partition matrix and cluster prototypes of every sub data sets are computed by FuzzyC-Means(FCM)algorithm. Secondly, the number of nearest neighbors is setting and density of the cluster prototypes isgained by K-nearest neighbor method, forming density matrix. Thirdly, it dynamically adjusts collaborative coefficient bythe correlation of density matrix. Lastly, it clusters objects with dynamical collaborative coefficient. Examples are providedto demonstrate the rationality of collaboration coefficient and the performance of collaboration FCM algorithm.