基于MCD的稳健主成分算法及其实证分析  

A Robust Principal Component Analysis Based on MCD Estimator and Its Empirical Study

作　　者：王斌会[1] 陈一非[1] 

机构地区：[1]暨南大学统计系,广东广州510632

出　　处：《数理统计与管理》2006年第4期462-468,共7页Journal of Applied Statistics and Management

基　　金：广东省科技计划攻关项目(编号:2004B10101010)

摘　　要：主成分分析方法是在经济管理中经常使用的多元统计分析方法,在变量降维方面扮演着很重要的角色,是进行多变量综合评价的有力工具。但传统的主成分分析对于异常值十分敏感,计算结果很容易受到异常值影响,而实际数据常包含异常情况,通常分析很少考虑它们的作用。本文基于MCD估计提出一种稳健的主成分分析方法,模拟和实证分析结果表明,该方法对于抵抗异常值有很好的效果。Principal component analysis（PCA）is a frequently used muhivariable analysis method in economics and management, it plays an important role in dimension reduction and is a powerful tool for overall evaluation. But traditional PCA is very sensitive to outliers and the results are easily affected by them. Real-life data always include abnormal situations which is usually lack of consideration. A robust PCA based on MCD estmator is put forward in this paper. Simulations and empirical study prove that it is very effective in resistance of outliers.

关 键 词：异常值 MCD估计 主成分分析 稳健主成分分析 

分 类 号：O212[理学—概率论与数理统计]