An Integrated Approach to Outlier Identification and Variable Selection in Discriminant Analysis
An important problem in discriminant analysis that has received little attention in the literature is the effect of outliers when variable selection forms part of the analysis. In this paper we argue that variable selection and outlier identification should not be done sequentially, but should rather be integrated. We investigate an integrated approach, and compare its classification performance to that of a sequential approach in a limited simulation study. In the configurations studied, we find that none of the approaches consistently outperforms the others.