主成分回归中异常点的稳健诊断  

Robust Diagnostic Method of Outliers in Principal Component Regression

作　　者：郭东星[1] 刘伟新[1] 

机构地区：[1]山西医科大学卫生统计教研室,030001

出　　处：《中国卫生统计》2008年第1期31-34,共4页Chinese Journal of Health Statistics

基　　金：山西省自然基金资助项目(项目编号20021104)

摘　　要：目的探讨主成分回归模型中存在异常点时的稳健诊断方法。方法采用基于MVT和LMS方法的一种稳健主成分回归方法来诊断异常点,并结合实例进行方法的对比。结果与未去除异常点时得到的回归方程相比较,具有较小的残差平方和,拟合效果较好。解决了主成分回归中存在异常点的问题。结论当主成分回归中存在异常点时,本文中所述的稳健诊断方法具有较高的稳健性,在主成分回归分析中有较好的应用前景。Objective To explore the robust diagnostic method when there are outliers in principal component regression. Methods We introduce a robust principal component regression based on MVT and LMS to detect outliers, and compare methods using a practical example. Results The model of the data in which outliers is detected and removed has a lower PRESS and a better fitness than that of the original data It is better to solve the outliers problem in PCR. Conclusion The robust diagnostic method in principal component regression has high robustness when there are outliers in PCR and should be widely used.

关 键 词：MVT方法 LMS估计 稳健主成分回归 异常点诊断 

分 类 号：R195[医药卫生—卫生统计学]