基于相关分析的粗差可区分性

Distinguishability of Outlier Based on Correlative Analysis

提出了应用偏相关系数来区分多维粗差和用复相关系数对多维粗差总体显著性进行检验并定位的方法。反复计算这些相关系数并进行检验,可以相对正确地进行多维相关观测的粗差定位。

This paper is the sequential research of the “correspondence based outlier analysis”. For correlative observations, the partial correlation coefficients are used to distinguish the multidimensional outliers, and the compound correlation coefficients are used to proof-test the holistic prominence of multidimensional outliers. By computing the correlation coefficient time and again, the authors locate the outliers of multidimensional outliers more accurately.