Zero-mean difference discrimination and the absolute linear discriminant function

Abstract

A simple model for discriminating equal mean data is to perform a linear discriminant analysis on the absolute value of deviations from the mean. This avoids the necessity of writing a program to calculate s quadratic discriminant function, and it also seems to have some desirable robustness properties when long-tailed contamination is present. The method is applied to the well-known Stocks twins data.