Table 3.

Percentage of Misclassification in Discriminant Analysis by Data Set and Method Used

Data set	Parametric			NNET	Nonparametric			RM
LDA	LOG	QDAa	KER	1NN	3NN
Alu21	40	32	NA	38	37	32	36	30
Alu8	49	35	51	35	49	42	38	37
β-globin	58	63	65	52	58	52	53	50
Y98	79	71	70	27	61	27	27	40
Y99	83	81	73	27	43	27	27	37

^aThe QDA method could not be used for the Alu21 data set because of the insufficient number of individuals with the complete, 21-locus genotype in some samples.(LDA) Linear discriminant analysis; (LOG) logistic discriminant analysis; (QDA) quadratic discriminant analysis; (NNET) neural networks; (KER) nonparametric discriminant analysis with Gaussian kernel; (1NN and 3NN) nonparametric discriminant analysis with, respectively, 1 and 3 nearest neighbors; (RM) Rannala and Mountain (1997) method.