Table 4 Best overall models, which are based on the highest ratio between the root mean squared error (RMSE) and squared spearman correlation (rho²), identified for each predictive scenario and the respective hyperparameters used in the model

mRFI_RFI

xgboost

max_depth = 5; eta = 0.05; nround = 315; min_child_weight = 1; subsample = 1; colsample_bytree = 0.5

0.10

0.86

mFCR_FCR

deeplearning

hidden = 5,5,5,5,5; epochs = 5.087879; l1 = 1e-04; rate = 1

0.19

0.73

mCons_RFI

deeplearning

hidden = 100,100,100; epochs = 51.294697; l1 = 1e-05; rate = 0.001

0.07

0.37

mCons_FCR

xgboost

max_depth = 5; eta = 0.05; nround = 343; min_child_weight = 1; subsample = 1; colsample_bytree = 0.5

0.17

0.93

mRFI_RFI + VARs

xgboost

max_depth = 5; eta = 0.05; nround = 303; min_child_weight = 1; subsample = 1; colsample_bytree = 0.5

0.12

0.73

mFCR_FCR + VARs

RF

mtry = 80; ntrees = 27; sample_rate = 1.0

0.19

0.51

mCons_RFI + VARs

RF

mtry = 45; ntrees = 15; sample_rate = 0.5

0.07

0.62

mCons_FCR + VARs

deeplearning

hidden = 10,10,10,10; epochs = 10.450117; l1 = 1e-04; rate = 1

0.23

0.80