秩变换判别方法对异常点的稳健性研究  

Robustness of The Discriminant Procedure Using Ranks For Outliers

作　　者：饶克勤[1] 黄湘宁[2] 

机构地区：[1]卫生部卫生统计信息中心 [2]北京医科大学卫生统计与医学人口教研

出　　处：《中国卫生统计》1995年第3期1-3,共3页Chinese Journal of Health Statistics

摘　　要：Fisher 线性判别方法的判别效率可能因样本中异常点的存在而降低。本文考察了秩变换判别方法对异常点的稳健性。对均值平移和方差膨胀两类重要类型的异常点,计算机模拟表明秩变换判别方法较 Fisher 线性判别方法更稳健。Considering that outliers decline the efficiency of Fisher's linear discriminant procedure (FLDP) and that rank-trans- forms are robust for outliers,the robustness of the disciminant proceduer using ranks(RLDP)for outliers is investigated.The simulation results based on the outliers which come from the means-shift and variance-inflation show that discriminant re- sults of RLDP are better than ones of FLDP,which means that the former is robuster that later for outliers.

关 键 词：秩变换判别 异常点 稳健性 医用数理统计 FLDP 

分 类 号：R195.1[医药卫生—卫生统计学]