Neural Discriminant Models, Bootstrapping, and Simulation

Table 4

Comparison of various discriminant methods for a thyroid disease database.

Model	Apparent error rate
	Training sample	Test sample
Multiple-group neural discriminant model (ℎ = 2)	0.009	0.017
Multiple-group logistic discrimination with linear effects of covariates	0.028	0.041
Multiple-group logistic discrimination with linear + quadratic effects of covariates	0.019	0.032
Tree-based model	0.002	0.073
2-nearest neighbor	0.017	0.074
3-nearest neighbor	0.038	0.089
Kernel smoother	0.054	0.062
SVM	0.045	0.052
Proportional odds model	0.048	0.055
VGAM	0.000	0.021