基于回归分析的全体与类样本分类器的比较研究

COMPARATIVE STUDY ON LINEAR REGRESSION BASED POPULATION AND CLASS SAMPLE CLASSIFIER

针对基于L1范数的Lasso回归与基于L2范数的Ridge回归模型,分别讨论两种分类器的设计方法,即基于Lasso回归的全体与类样本分类器和基于Ridge回归的全体与类样本分类器。分别在2个大样本数据库与2个小样本数据库对所给出的方法进行比较研究与分析,结果表明基于全体样本的分类器更适合小样本问题,而基于类样本的分类器更适合大样本问题。

By examining Ll-paradigm based Lasso regression and L2-paradigm based Ridge regression models, the design methods of two classifiers, i. e. , Lasso regression based population and class sample classifier, and Ridge regression based population and class sample classifier, are separately discussed. Comparative study and analysis are carried out separately upon 2 large sample size databases and 2 small sample size databases. Results show that population sample based classifier suits better for small sample size problems whereas class sample based classifier suits better for large sample size problems.