An Alternative Method of Detecting Outlier in Multivariate Data using Covariance Matrix
In the Multivariate data analysis, the detection of outliers is important and necessary though this may be difficult and can pose a problem to the analyst. When a set of data is contaminated, the values obtained from such set of data are distorted and the results meaningless. In this work we present a simple multivariate outlier detection procedure using a robust estimator for variance-covariance matrix by using the best units from the available data set that satisfied the three predetermined optimality criteria, selected from all possible combinations of sub-sample obtained. The proposed estimator used is the variance-covariance estimator of the best unit multiplied by a constant. It is observed that, the proposed method combined the efficiencies of the classical and the existing robust (MCD and MVE) of being able to signal when there are few and multiple outliers in multivariate data.
