剔除异常数据的稳健性处理方法  

The robustness data processing method about eliminating outlier

作　　者：林洪桦[1] 

机构地区：[1]北京理工大学,北京100081

出　　处：《中国计量学院学报》2004年第1期20-24,共5页Journal of China Jiliang University

摘　　要：　由于以往的剔除异常数据的方法,包括国家标准GB8056-87和GB6380-86中所给出的判别异常数据的方法,多利用残差和标准差且基于假定某种典型概率分布,故存在其局限性.再者,在初始判别时就可能受异常数据的影响,尤其在小样本下会有误判.经作者各种各样的仿真试验,提出一些剔除异常数据的稳健性通用处理方法,以期适于在不同概率分布的小样本下应用.Because the methods to eliminate outlier in the past, including the criterion of outlier given in the national standard GB8056-87 and GB6380-86, which often utilized residuals and standard deviation, and was based on a certain typical probability distribution assumption, there was their limitation on it. And these methods may be affected by outlier at the beginning of discrimination, especially in small sample situation misjudge was occurred. In this paper, a universal robustness method to eliminate outlier was proposed according to kinds of simulation experiments by the author, and expected to be adopted in small sample data of variant probability distribution.

关 键 词：异常数据 稳健性 β分布 粗大误差 剔除方法 

分 类 号：O212[理学—概率论与数理统计]