PCM算法存在聚类重叠的缺陷,PFCM算法同时利用隶属度与典型值把数据样本划分到不同的类中,提高了算法的抗噪能力,但PFCM算法对样本分布不均衡的聚类效果并不十分理想。针对此不足,可以通过Mercer核把原来的数据空间映射到特征空间,并为特征空间的每个向量分配一个动态权值,从而得到特征空间内的目标函数。理论分析和实验结果表明,相对于其他经典模糊聚类算法,新算法具有更好的健壮性和聚类效果。
PCM algorithm often tends to find the identical cluster.Proposed PFCM,which divides the data set into different clusters through producing memberships and possibilities simultaneously,along with the cluster centers.But when two highly imbalanced samples clusters are given,PFCM fails to give the desired results.In order to overcome the weakness,this paper firstly mapped the original data space to a high-dimensional feature space by Mercer kernel functions,and assigned an addtional weighting factor to each vector in the feature space.Then introduced a modified objective function for fuzzy clustering in the feature space.Theoretical analysis and experimented results testify that the new algorithm has more robust and higher clustering accuracy compared with those classic fuzzy clustering algorithm.