Table 5.

Average number of models in phases 1 and 2 over which averaging is carried out by the ISMB and NISMB algorithms. Both algorithms average over the same number of models in each phase. Both algorithms select the same models in phase 1 but potentially different models in phase 2. The number of models in phases 1 and 2 is the sum of the models selected in the two phases.

Data Set	# models phase 1	# models phase 2	# models phases 1 and 2
australian	28.55	11.00	39.55
breast-cancer	18.85	10.15	29.00
cleveland	20.45	11.99	32.44
corral	10.65	15.03	25.68
crx	32.10	13.42	45.52
diabetes	11.65	10.03	21.68
flare	20.75	11.44	32.19
german	22.45	19.23	41.68
glass2	12.05	13.26	25.31
glass	15.80	10.73	26.53
heart	18.50	11.32	29.82
hepatitis	27.45	26.63	54.08
iris	7.25	10.74	17.99
lymphography	51.55	37.83	89.38
pima	40.40	16.97	57.37
postoperative	12.00	10.02	22.02
sonar	11.65	10.09	21.74
vehicle	1.15	21.09	22.24
vote	59.80	18.44	78.24
wine	39.30	10.73	50.03
zoo	45.55	13.53	59.08