A comparative study of methods to handle outliers in multivariate data analysis

Abstract

Detecting outliers is an integral part of data analysis that sheds light on points that do not conform with the rest of the data. Whereas in univariate data, outliers appear at the extremes of the ordered sample, in the multivariate case they may be defined in many ways and are not generally based on an assumed statistical model. We present here methods for detecting multivariate outliers based on various definitions and illustrate their features by applying them to two sets of data. No single approach can be recommended over others, since each one aims at detecting outliers of a particular kind.