Table 1.

Performance of BRT models within the 5-fold gridded cross validation structure using different combinations of meta-parameters: tree complexity (tc), learning rate (lr), and bagging fraction (bf).

Tree complexity	Learning rate	Bagging fraction	No. trees	mean deviance	MSE
5	0.005	0.5	850	2.96	4.68
5	0.005	0.6	950	2.68	4.38
5	0.005	0.7	1200	2.62	4.35
5	0.001	0.5	4450	2.79	4.46
5	0.001	0.6	5000	2.74	4.42
5	0.001	0.7	5200	2.81	4.55
5	0.0005	0.5	8850	3.13	4.90
5	0.0005	0.6	10000	3.06	4.84
5	0.0005	0.7	9150	3.06	4.88
7	0.005	0.5	650	2.78	4.40
7	0.005	0.6	950	2.55	4.18
7	0.005	0.7	700	2.67	4.33
7	0.001	0.5	3200	2.74	4.36
7	0.001	0.6	4400	2.68	4.30
7	0.001	0.7	4000	2.61	4.24
7	0.0005	0.5	5600	2.95	4.60
7	0.0005	0.6	8200	2.72	4.34
7	0.0005	0.7	7200	2.70	4.34
9	0.005	0.5	350	3.03	4.68
9	0.005	0.6	600	2.68	4.28
9	0.005	0.7	650	2.51	4.09
9	0.001	0.5	2200	2.83	4.42
9	0.001	0.6	3400	2.54	4.11
9	0.001	0.7	3600	2.54	4.13
9	0.0005	0.5	4550	2.96	4.58
9	0.0005	0.6	6750	2.59	4.15
9	0.0005	0.7	7000	2.57	4.16

Performance measures reported are mean Poisson deviance and mean squared error (MSE) of the model across the 5-folds. The meta-parameter combination with the lowest mean deviance and MSE, and >1000 fitted trees is shown in bold.