Table 2.

Optimal parameter settings used in calibrating the BRTs that produced the best performing introduction–naturalization–invasion models. To reduce overfitting, we used cross-validation that was performed by splitting 75 % of the data for training the model and 25 % for testing. We tested various learning rates (0.1–0.0005), bag fractions (0.1–0.8) and levels of tree complexity (1–5). By trial and error, we determined the most effective algorithm parameters for our dataset, which is depicted below.

	Introduction model	Naturalization model	Invasion model
Sample size (n)
Full dataset	3494	514	46
Training data	2621	386	–
Test data	873	128	–
Parameters
Learning rate	0.001	0.001	0.001
Tree complexity	3	3	3
Bag fraction	0.5	0.5	0.75