一种用于二类样本判别分析的PLS方法

A New PLS Method Suitable to Deal with the Discriminant Analysis Problems with Two Class Samples

提出了一种新的用于两类样本判别分析问题的PLS方法，该法对响应函数y作了类似神经网络算法中用的Sigmoid函数转换，可用一种新的优化目标判据来提取一组PLS方法中两两正交的隐变量t1，t2......，用这些变量可构成判别分类图，并可得到比较理想的判别方向矢量．

A new PLS method that is very suitable dealing with the discriminant analysis problems with two class samples has been developed. In this method the dependent function y has been transformed by use of the Sigmoid function, that is similar to the neural network algorithm. Therefore a new optimization criteria can be used to extract the latent variables t1, t2, .. .., in PLS method. The good classification graphics constructed by the latent variables, and the ideal direction vectors of discrimination can all be obtained.