Table 1.

Prediction error rates and computational time in seconds for the simulated examples.

MSVM Method	Ex 1 Linear			Ex 2 Poly			Ex 3 Linear
	Error	Time	p-value	Error	Time	p-value	Error	Time	p-value
MSVM2	14.67	18	0.000	15.53	54	1.000	14.48	306	0.419
MSVM3	15.21	21	0.000	14.96	48	1.000	14.41	299	0.533
MSVM4	22.47	20	0.000	21.79	64	0.000	14.56	338	0.297
MSVM5	11.34	151	0.000	17.76	452	0.000	-	-	-
MSVM6	14.98	27	0.000	16.14	77	0.000	14.76	422	0.088
RAMSVM	9.80	13	-	15.71	23	-	14.43	115	-
MSVM Method	Ex 1 Gauss			Ex 2 Gauss			Ex 3 Poly
	Error	Time	p-value	Error	Time	p-value	Error	Time	p-value
MSVM2	8.64	13	1.000	11.62	17	0.000	14.64	599	0.109
MSVM3	9.16	11	0.000	12.88	21	0.000	14.91	478	0.010
MSVM4	11.71	15	0.000	15.19	23	0.000	14.88	705	0.013
MSVM5	14.09	277	0.000	15.82	298	0.000	-	-	-
MSVM6	11.57	15	0.000	13.57	30	0.000	14.29	818	0.581
RAMSVM	8.78	14	-	11.34	21	-	14.34	355	-

Poly: Second order polynomial kernel learning. Gauss: Gaussian kernel learning. The standard errors of the error rates range from 0.6% to 1.7%. The standard errors of the computational time range from 1 to 32 seconds. Note that MSVM5 cannot be computed for Example 3 due to its large n.