Table 2.

Hessleskew and Hagg fields results in cross-validation, laboratory and on-line predictions using local, and spiked European dataset based on gradient boosted machines (GBM), artificial neural networks (ANNs) and random forests (RF) models.

Hessleskew									Hagg
			Local			European					Local			European
			RMSE	R2	RPD	RMSE	R2	RPD			RMSE	R2	RPD	RMSE	R2	RPD
GBM	n.trees								n.trees
Cross-	100	TN	0.01	0.63	1.64	0.01	0.96	4.81	100	TN	0.02	0.63	1.66	0.01	0.96	5.01
	100	TC	0.16	0.67	1.75	0.06	0.98	6.48	100	TC	0.19	0.65	1.70	0.12	0.98	6.49
Lab Prediction	100	TN	0.01	0.60	1.60	0.02	0.87	2.79	100	TN	0.02	0.62	1.65	0.02	0.79	2.21
	100	TC	0.23	0.60	1.59	0.20	0.82	2.40	100	TC	0.21	0.61	1.61	0.19	0.83	3.03
On-line Prediction	100	TN	0.02	0.53	1.48	0.02	0.66	1.80	100	TN	0.02	0.59	1.58	0.02	0.77	2.11
	100	TC	0.26	0.54	1.49	0.24	0.66	1.78	100	TC	0.22	0.52	1.46	0.20	0.79	2.95
ANN	size								size
Cross- validation	2	TN	0.01	0.62	1.62	0.01	0.77	2.08	2	TN	0.03	0.35	1.25	0.03	0.73	1.92
	2	TC	0.21	0.44	1.34	0.15	0.86	2.69	2	TC	0.18	0.70	1.82	0.18	0.86	2.79
Lab Prediction	2	TN	0.01	0.69	1.81	0.01	0.71	2.02	2	TN	0.02	0.66	1.74	0.02	0.68	1.87
	2	TC	0.25	0.51	1.45	0.20	0.83	2.44	2	TC	0.25	0.47	1.40	0.21	0.84	2.59
On-line Prediction	2	TN	0.02	0.26	1.18	0.01	0.68	1.78	2	TN	0.04	0.11	1.07	0.03	0.59	1.59
	2	TC	0.34	0.19	1.13	0.25	0.78	2.14	2	TC	0.27	0.75	1. 95	0.20	0.85	2.63
RF	ntree								ntree
Cross- validation	100	TN	0.01	0.83	2.45	0.01	0.96	4.83	100	TN	0.01	0.84	2.50	0.01	0.97	5.58
	100	TC	0.12	0.82	2.38	0.06	0.98	6.48	100	TC	0.13	0.84	2.52	0.10	0.98	7.54
Lab Prediction	100	TN	0.01	0.82	2.40	0.02	0.81	2.33	100	TN	0.02	0.78	2.18	0.02	0.84	2.51
	100	TC	0.25	0.75	2.02	0.23	0.78	2.16	100	TC	0.18	0.81	2.35	0.14	0.88	3.49
On-line Prediction	100	TN	0.01	0.72	1.93	0.04	0.55	1.52	100	TN	0.02	0.79	2. 20	0.02	0.83	2.40
	100	TC	0.21	0.69	1.82	0.20	0.75	2.13	100	TC	0.15	0.77	2.13	0.14	0.86	3.24

n.trees = total number of trees to fit., size = number of units in the hidden layer, ntree = number of trees.