DISCRIMINATION BY MEANS OF COMPONENTS THAT ARE ORTHOGONAL IN THE DATASPACE

Krzanowski (J. Chemometrics, 9, 509 (1995)) proposed a method for obta ining so-called orthogonal canonical variates (henceforth called compo nents) for discrimination purposes. In contrast with ordinary discrimi nant analysis, this method employs components that are orthogonal in t he original data space. These components are derived in a successive w ay, thus optimizing discrimination of a component given the previously extracted components. Two alternative procedures are proposed to extr act the desired number of components simultaneously, yielding a better overall discrimination. The simultaneous approaches are applied to th e same two data sets as analysed by Krzanowski, as well as to Anderson 's Iris data, and a comparison of discriminatory quality of the soluti ons is presented. (C) 1997 John Wiley & Sons, Ltd.