一种新的分类方法

本文在属性聚类网络的基础上 ,提出了堆近邻分类方法 .通过将无监督的属性聚类加上有监督信息 ,能自适应地优选堆数 .样本所考察的近邻个数依据它所在的堆的大小 ,因而每个样本所考查的近邻的个数不是完全相等的 .这种方法可用到高维小样本的数据分类问题中 .我们将它应用到基因表达谱形式的癌症辩识问题中 ,结果表明分类性能得到了较大的提高

An New Classification Method

In this paper, we present the cluster neighborh ood classification method based on attribute clustering network-an unsupervised clustering. The number of the clusters is decided adaptively by using unsupervi sed clustering and supervised information. The number of training samples of one sample′s considered neighbor is according to the size of the cluster to which the sample belongs. This method can be used in high dimension and small size dat a set. We use it in a question concerned with gene expression data analysis to d iscriminate tumor from normal tissues, and get good performance.