Variable Selection Procedure for Discrimination Between two Multinormal Populations with Common Dispersion Matrix Proportional to a Known Positive Definite Matrix

Abstract

Abtsrcat

A stepwise procedure is described for selecting a suitable sub-set of characters for discriminating between two multivariate normal populations with a common dispersion matrix which is an unknown scalar multiple of a known positive denite positive denite matrix. The procedure assumes with a prescribed condence coecient that the distance based on the subset of selected variables is not less than a prescribed proportion of the distance based on all variables. The procedure is consistent in natural sense.

Keywords

Discriminant analysis Mahalanobis distance Confidence bound Consistency Steepest step acceptable set.

Get full access to this article

View all access options for this article.