Performance of statistical and machine learning-based methods for predicting biogeographical patterns of fungal productivity in forest ecosystems

Forest Ecosystems

Table 1 Hyperparameters ranges and types for each machine learning model

Algorithm	Hyperparameter ID	Type	Lower	Upper
RF	mtry	Integer	1	N° of predictors
	min.node.size	Integer	1	100
	num.trees	Integer	2	500
XGB	nrounds	Integer	1	100
	gamma	Numeric	1	25
	max_depth	Integer	1	15
	eta	Numeric	0.1	1
SVM	cost	Numeric	1	50
SVM	gamma	Numeric	0.1	1
ANN	epochs	Integer	1	100
ANN	batch_size	Integer	1	N° observations

“Hyperparameter id” corresponds to the names specified in the R package used to train each model. RF (random forest), XGB (extreme gradient boosting), SVM (support vector machine), and ANN (artificial neural network) models were trained using the R packages “ranger” (v0.12.1; Wright and Ziegler 2017), “xgboost” (v0.90.0.2; Chen et al. 2019), “e1071” (v1.7–2; Meyer et al. 2019b) and “keras” (v2.3.0.0.0; Allaire and Chollet 2019), respectively