摘要
k-近邻分类是一种流行且成功的非参数分类方法,但其分类性能由于离群点的存在而受到损害.为克服离群点对分类性能的不利影响,提出了一个k-近邻分类的变形和一个基于局部均值向量与类均值向量的近邻分类方法.该方法利用了未分类样本在每个训练类中k个近邻的局部均值的信息和整体均值的知识,不仅能够克服离群点对分类性能的影响,而且取得了比传统的k-近邻分类一致好的分类性能.
The k-nearest neighbour classification is a very popular and successful nonparametric classification method, but its classification performance usually suffers from the existing outliers. To overcome the adverse effect of the existing outliers on classification performance, a variant of the k-nearest neighbour classification and a nearest neighbour classification method based on the local mean and class mean are proposed. The information of the local mean of the k nearest neighbours of the unclassified sample in each class and the knowledge of the ensemble mean are taken into account in the classification method. The proposed classification method overcomes the influence of the existing outliers and achieves a uniformly good classification performance compared with the traditional k-nearest neighbour classification.
出处
《控制与决策》
EI
CSCD
北大核心
2009年第4期547-550,556,共5页
Control and Decision
基金
国家973计划项目(2004CB720703)
关键词
k一近邻分类
局部均值
类均值
交叉验证
k-nearest neighbour classification
Local mean
Class mean
Cross-validation