Table 3.

Performance of the ANNs as a function of the number of neurons in the hidden layer.

		Training Set					Test Set
NN	Score	r²	MAE	RMSE	RAE	RRSE	q²	MAE	RMSE	RAE	RRSE
1	-	0.51	0.86	1.03	87	68	-	-	-	-	-
2	-	0.72	0.69	0.85	69	72	-	-	-	-	-
3	-	0.75	0.69	0.86	69	72	-	-	-	-	-
4	-	0.77	0.68	0.84	68	70	-	-	-	-	-
5	0.78	0.78	0.64	0.81	65	68	0.80	0.64	0.82	63	66
6#	0.80	0.79	0.65	0.82	65	68	0.85	0.60	0.75	59	61
7	0.81	0.81	0.57	0.73	56	62	0.81	0.59	0.76	58	62
8	0.76	0.79	0.68	0.85	68	71	0.77	0.69	0.90	68	73
9	0.80	0.80	0.65	0.82	65	68	0.82	0.64	0.81	63	66
10	0.77	0.82	0.58	0.75	58	63	0.79	0.62	0.83	61	67

Note: NN = number of neurons in the hidden layer; r² = correlation coefficient for the training set; q² = correlation coefficient for the test set (r²_pred); RMSE = root mean square error; MAE = mean absolute error; RAE = relative absolute error; RRSE = root relative squared error; score = (1 − |(r² − q²)|) × q². ^#Standard NN.