Kvazikanonička diskriminativna analiza u dualnoj eta metrici

Abstract

Predložena je jedna nova metoda kvazikanoničke diskriminativne analize u kojoj se na Eta metriku reparametrizira skup kvantitativnih varijabli koje definišu prostor u kome će se formirati diskriminativne funkcije, i zatim maksimiziraju kovarijanse tako reparametriziranih varijabli sa tim varijablama u standardnoj I metrici koje su projektovane u hiperkub koga tvore binarni vektori formirani opisom skupa objekata na nekoj nominalnoj varijabli. Definisana su formalna svojstva ove metode, napisan jedan program kojim je ta metoda implementirana tako da se može izvesti u standardnom SPSS okruženju, i naveden jedan numerički primer ponašanja tog programa.

The paper proposes one method of quasi-canonical discriminative analysis, in which a set of quantitative variables, which define the space in which discriminative functions will be formed, are reparameterised and then the covariance of such reparameterised variables are maximized with these variables in standard I metrics which are projected in hyper-cube composed of binary vectors formed by the set of objects circumscribed on some nominal variable. Formal characteristics of this method are defined, one program, by which this method has been implemented was written, so that it can be realized in standard SPSS environment, and one numeric example of this program's behavior is presented.