Table 2.

Best supervised learning models.

Modela	Resampling	F1 score	AUCb	Accuracy	Recall	Precision
Random forestc	SMOTEd	0.186	0.591	0.689	0.476	0.116
Extreme gradient boostinge	Random oversampling	0.179	0.591	0.614	0.563	0.106
K-nearest neighborsf	Random undersampling	0.181	0.605	0.541	0.682	0.105
Naïve Bayesg	SMOTE	0.184	0.602	0.596	0.609	0.108
Logistic regression	SMOTE	0.185	0.608	0.570	0.654	0.108

^aAll models except neural network applied to known predictors only.

^bAUC: area under the receiver operator characteristic curve.

^cRandom forest hyperparameters: estimators=200, maximum features=8, maximum leaf nodes=300.

^dSMOTE: synthetic minority oversampling technique.

^eExtreme gradient boosting hyperparameters: booster=gbtree, η=0.9, γ=0, α=1, λ=0.

^fK-nearest neighbors: N=500.

^gNaïve Bayes: α=0.