hrvatski

Algoritam za diskriminativnu analizu u Mahalanobisovom prostoru

Predloženi algoritam se razlikuje od standardnih algoritama za kanoničku diskriminativnu analizu u nekoliko bitnih karakteristika. Razlike su u tome što se diskriminativni koeficijenti i diskriminativni faktori određuju i metrici standardiziranih varijabli i u metrici varijabli transformiranih u Mahalanobisov oblik, pa zatim u tome što se značajni diskriminativni faktori (značajnost se određuje prema informacijskom doprinosu faktora objašnjenju ukupne intergrupne varijance) rotiraju u položaj koji maksimizira normalizirani quartimax kriterija. Centroidi grupa i za nerotirane i za rotirane diskriminativne varijable određuju se na koordinatnom sistemu i ishodištem u centroidu ukupnog uzorka, statistički se provjerava značajnost diskriminativnih funkcija pomoću X2 testa, te se rezultati X2 testa provjeravaju se jednofaktorskom, univarijatnom analizom varijance za svaku zadržanu nerotiranu i rotiranu diskriminativnu funkciju.

kanonoička diskrimiantivna analiza ; Mahalanobisov prostor

The algorithm was proposed and the program written for canonical discriminative analysis of group of variables with non-singular matrix of covaiances, transformed to the Mahalanobis form. The algorithm differs from standard algorithms for discriminative analysis ; in that it determines the discriminative coefficients and discriminative factors in both the metrics of standardized and in the metrics of variables transformed to Mahalanobis form, in that it rotates the significant discriminative factors in to a position which maximizes the normalized quartimax criterion in which the number of retained factors is determined according to the informational contribution of factors to the explanation of entire intergroup variance ; in that it determines the group centroides for both the unrotated and rotated discriminative variables in the coordinate system with the starting point in the centroid of the entire sample ; in that it statistacaly tests significance of discriminative functions retained according to the informational criterion by means fo Chi square criterion (Rao, 1951) ; what is additionally tested with univariate analysis of