Table 3.

Average misclassification error rates for optimal designs obtained by tree classification (cross-validated), random forest classification (using out-of-bag class predictions), and ABC approaches under the 0–1 loss (01L) or multinomial deviance loss (MDL) as well as for the equidistant designs for the infectious disease example. The average misclassification error rates were calculated by repeating the random forest classification procedure 100 times (see text) and taking the average. The standard deviations are given in parentheses

Design	$n=1$	$n=2$	$n=3$	$n=4$	$n=5$
Tree 01L	0.5554	0.5158	0.5133	0.5116	0.5129
	(0.0023)	(0.0024)	(0.0026)	(0.0022)	(0.0025)
RF 01L	0.5548	0.5160	0.5132	0.5113	0.5138
	(0.0027)	(0.0025)	(0.0025)	(0.0025)	(0.0025)
ABC 01L	0.5547	0.5161	0.5196	0.5046	0.5339
	(0.0023)	(0.0025)	(0.0030)	(0.0024)	(0.0027)
Tree MDL	0.5547	0.5178	0.5152	0.5183	0.5159
	(0.0023)	(0.0030)	(0.0028)	(0.0028)	(0.0025)
RF MDL	0.5550	0.5179	0.5118	0.5104	0.5128
	(0.0022)	(0.0026)	(0.0026)	(0.0026)	(0.0026)
ABC MDL	0.5553	0.5221	0.5216	0.5226	0.5416
	(0.0020)	(0.0028)	(0.0028)	(0.0029)	(0.0025)
Equidistant	0.6592	0.6200	0.5760	0.5537	0.5519
	(0.0029)	(0.0025)	(0.0027)	(0.0029)	(0.0032)