Table 1.

Results of the performance of the linear regression algorithms, the multilayer perceptron (MLP), and the support vector regressor (SVR) for Brazil's data set.

		Correlation coefficient		RMSE (%)		Training time (s)
Regression method	Configuration	Average	Standard deviation	Average	Standard deviation	Average	Standard deviation
Linear regression	–	0.9826	0.143	11.42	22.94	0.003	0.006
	10 trees	0.9737	0.072	22.35	24.08	0.06	0.01
	20 trees	0.9752	0.070	21.59	23.03	0.20	0.04
	30 trees	0.9756	0.069	21.32	22.68	0.29	0.06
Random forest	40 trees	0.9759	0.068	21.17	22.33	0.22	0.04
	20 neurons	0.9948	0.006	11.29	7.12	13.91	0.19
MLP	30 neurons	0.9867	0.011	17.9	4.12	6.48	0.30
	Polynomial kernel, p = 1	0.9670	0.109	18.39	24.08	3.82	2.14
	Polynomial kernel, p = 2	0.9818	0.006	31.70	10.62	7.99	10.57
	Polynomial kernel, p = 3	0.9623	0.004	45.15	17.35	7.55	9.45
SVR	RBF kernel	0.5341	0.375	87.19	14.94	39.97	56.17

For each regressor, we calculated the correlation coefficient, the relative square error [RMSE (%)], and the training time.