多元数据离群点探测的倾斜重加权方法

Multivariate Outlier Detection Based on Tilting Minimum Covariance Determinant Method

在对数据的处理中,关于多元数据的离群点的探测日益受到重视。然而现有的方法存在着不够稳健、准确率较低等缺点。本文提出了两种多元数据离群点探测方法,并在重加权最小协方差行列式方法的基础上,采用更为科学的倾斜重加权方法,计算出更为稳健的估计量,从而达到更加有效准确地探测离群点的目的。

Multivariate outlier detection attracts more and more interests in data processing nowadays. However, the existing methods are often not robust or accurate enough. This paper summarizes two kinds of existing met hods, and proposes a tilting method based on Reweighted Minimum Covariance Det erminant. Using the Tilting Minimum Covariance Det erminant, we can get more rob ust estima tors so as to achieve more effective and accurate detection of outliers.