高维少样本数据的特征压缩
Feature reduction on high-dimensional small-sample data
摘要
针对一类高维少样本数据的特点,给出了广义小样本概念,对广义小样本进行信息特征压缩:特征提取(降维)和特征选择(选维)。首先介绍基于主成分分析(PCA)的无监督与基于偏最小二乘(PlS)的有监督的特征提取方法;其次通过分析第一成分结构,提出基于PCA与PlS的新的全局特征选择方法,并进一步提出基于PlS的递归特征排除法(PlS-rfE);最后针对MITAMl/All的分类问题,实现基于PCA与PlS的特征选择和特征提取,以及PlS-rfE特征选择与比较,达到广义小样本信息特征压缩的目的。 In view of the characteristics of small sample and high dimensional data,Generalized Small Samples(GSS) is defined.It reduces information feature of GSS:feature extraction(dimensionality extraction) and feature selection(dimensionality selection).Firstly,unsupervised feature extraction based on Principal Component Analysis(PCA) and supervised feature extraction based on Partial Least Squares(PLS) are introduced.Secondly,analyzing the structure of first PC,it presents new global PCA-based and PLSbased feature selection approaches,in addition recursive feature elimination on PLS(PLS-RFE) is realized.Finally,the approaches are applied to the classification of MIT AML/ALL,it performs feature extraction on PCA and PLS,and feature selection compared with PLS-RFE.The information compression of GSS is realized.