基于核聚类的K-均值聚类

摘要

In order to process data sets with linear non-separability or complicated structure, the Kernel Clustering-based K-means Clustering (KCKC) method is proposed.The method clusters patterns in the original space and then maps them into the kernel space by Radial Basis Function (RBF) kernel so that the relationships between patterns are almost kept.The method is applied to the RBF-based Neural Network (RBFNN), the RBF-based Support Vector Machine (RBFSVM), and the Kernel Nearest Neighbor Classifier (KNNC).The results validate that the proposed method can generate more effective kernels, save the kernel generating time in the kernel space, avoid the sensitivity of setting the number of kernels, and improve the classification performance.%为处理线性不可分、结构复杂的数据集,提出基于核聚类的K-均值聚类(Kernel Clustering-based K-means Clustering,KCKC).该方法先在原始空间中对模式进行聚类,再由径向基函数(Radial Basis Function, RBF)核把它们映射到核空间,从而保持大部分模式之间的关系.把提出的方法应用到基于RBF的神经网络(RBF-based Neural Network,RBFNN)、基于RBF的支持向量机(RBF-based Support Vector Machine, RBFSVM)和核最近邻分类器(Kernel Nearest Neighbor Classifier,KNNC)中,结果表明本文提出的算法可以生成更有效的核,节省在核空间中的核生成时间,避免核数目设置的敏感性,并提高分类性能.