Logistic回归模型和判别分析方法的比较分析  

The Comparison of Logistic Regression Model and Discriminant Analysis Method

作　　者：冯珊[1] 方正武[2] 

机构地区：[1]中央财经大学统计学院,北京102206 [2]长江大学农学院,湖北荆州434025

出　　处：《长江大学学报（自科版）（上旬）》2014年第2期19-22,2,共4页JOURNAL OF YANGTZE UNIVERSITY (NATURAL SCIENCE EDITION) SCI ＆ ENG

基　　金：国家自然科学基金项目(31371529)

摘　　要：Logistic回归模型和判别分析方法都可以达到对样本分类的目的,比较和分析这2种方法的差异以及其各自的特点,可以为更好的应用提供参考。从类别表现、样本情况、归类函数、归类原则、预测功效等5个方面对Logistic回归模型中的基线-类别Logit模型和判别分析方法中具有代表性的Bayes后验概率判别、Fisher判别的基本思想和步骤进行比较,并通过"鸢尾花"样本数据的品种判别对这3种方法进行了实证分析。Logistic回归建立了总体各因子与总体类别的回归模型,在因子与类别之间形成了解释与被解释的关系;通过对回归系数的检验,可以探究各因子对总体类别的影响程度;为了取得更好的预测效果,在各种模型中,应尽量增大训练样本的容量。Logistic regression model and discriminant analysis method both can classify samples.This paper provides references for the better application of the two methods through comparing them and analyzing their characteristics.The basic ideas and steps of the baseline-category Logit model of Logistic regression model and the representative analysis methods of discriminant method-posteriori probability criterion and Fisher discriminant are compared from five aspects,such as category performances,sample situations,the classifying functions,the classification principles and prediction effects.Also the three discriminant methods are analyzed through example of the IRIS data sample.In terms of Logistic regression,the regression model between factors and totality is established and the relationship between them is explanatory and explained.By testing the regression coefficients,the influence degree of each factor on the totality can be explored.For better predictive validity,the sample size should be as large as possible.

关 键 词：LOGISTIC回归模型 判别分析方法 基线一类别Logit模型 Bayes后验概率判别 Fisher判别[ 

分 类 号：O212.2[理学—概率论与数理统计]