Table 1.

Performances of the five neural network models constructed via the 5-fold cross-validation

	E	Puniq	Rint	RBP	Punused
Model 1
Training sets	0.959 ± 0.012	0.612	0.461 ± 0.005	0.626 ± 0.158	0
Validation sets	0.960 ± 0.047	0.612	0.708 ± 0.013	0.289 ± 0.139	0
Test set	0.865	0.612	0.654	0.775	0
Model 2
Training sets	0.959 ± 0.010	0.709	0.417 ± 0.021	0.564 ± 0.084	0
Validation sets	0.960 ± 0.040	0.709	0.664 ± 0.021	0.207 ± 0.250	0
Test set	0.811	0.709	0.609	0.711	0
Model 3a
Training sets	0.967 ± 0.009	0.772	0.452 ± 0.006	0.815 ± 0.073	0.019
Validation sets	0.967 ± 0.035	0.772	0.689 ± 0.028	0.412 ± 0.292	0.019
Test set	0.946	0.772	0.648	0.895	0.019
Model 4
Training sets	0.951 ± 0.014	0.738	0.451 ± 0.011	0.741 ± 0.058	0
Validation sets	0.950 ± 0.054	0.738	0.709 ± 0.012	0.466 ± 0.490	0
Test set	0.811	0.738	0.668	0.725	0
Model 5
Training sets	0.976 ± 0.009	0.738	0.432 ± 0.007	0.825 ± 0.109	0
Validation sets	0.976 ± 0.035	0.738	0.696 ± 0.014	0.479 ± 0.292	0
Test set	0.865	0.738	0.654	0.781	0

^aModel 3 is adopted as the final model which is specified by values in bold face.