一种新的多类判别方法

A NEW METHOD OF MULTIPLE DISCRIMINANT ANALYSIS

本文提出一种非线性多类判别方法,即从G类已知正态总体的样本协方差矩阵出发,用主成份分析寻求各总体的超椭球近似域,以此为基础对未知样进行归类,并以欧氏距离作辅助判别指标,考察某些样品的不定解问题。此方法的主要特点是:不依赖样品归属的先验概率,避免主观性;适用于各总体协方差矩阵相同或不同、各类均值共线或不共线等多种情形;

A new method of multiple discriminant analysis was developed in this paper. At first, the hyperellipsoidal approximate domains of the G normal populations were found by using the idea of principal component analysis based on the variance-covariance matrices of G samples from the known normal populations,In addition,the euclidean distance was defined as an assistant discriminant index. Then, the classification of theunknown samples was discussed based on the hyperellipsoidal approximate domains and the assitant discriminant index. In the end,the method was used in some examples and the efficiency of the method and the problems associated with the efficiency were described.