Table 3.

Best combinations of scaling, pre-processing, and hyperparameters, shown from each classifier type. The upper table shows the 1-month VT recurrence endpoint, while the lower table shows the 1-year VT recurrence endpoint. RF: random forest; MLP: multilayer perceptron; XGB: extreme gradient boosting; SS: standard scaling; SMOTE: Synthetic Minority Oversampling Technique; VT: ventricular tachycardia; AUC: area under the curve.

1-month VT recurrence	Model	Pre-Processing	Oversampling	Hyperparameters	Mean AUC (Test)
	RF	Not used	Not used	RF__class_weight: None; RF__criterion: log_loss; RF__max_depth: 1; RF__max_features: 1; RF__min_samples_leaf: 2; RF__n_estimators: 300; RF__random_state: 0	0.730
	MLP	SS	SMOTE	NN__activation: logistic; NN__alpha: 0.00225; NN__hidden_layer_sizes: 3; NN__max_iter: 161; NN__random_state: 0; NN__solver: adam	0.729
	XGB	Not used	Not used	XGB__alpha: 7.99; XGB__colsample_bytree: 1.0; XGB__gamma: 0.1; XGB__max_depth: 3; XGB__min_child_weight: 0.0; XGB__random_state: 0; XGB__scale_pos_weight: 2.0; XGB__subsample: 0.5	0.708
1-year VT recurrence	Model	Pre-processing	Oversampling	Hyperparameters	mean AUC (test)
	RF	SS	SMOTE	RF__class_weight: balanced_subsample; RF__criterion: gini; RF__max_depth: 2; RF__max_features: 7; RF__min_samples_leaf: 2; RF__n_estimators: 320; RF__random_state: 100	0.713
	XGB	Not used	Not used	XGB__colsample_bytree: 1.0; XGB__gamma: 9.79; XGB__max_depth: 3; XGB__min_child_weight: 0.0; XGB__random_state: 0; XGB__scale_pos_weight: 3.14; XGB__subsample: 0.897	0.711
	MLP	SS	SMOTE	NN__activation: logistic; NN__alpha: 0.000192; NN__hidden_layer_sizes: 50; NN__max_iter: 50; NN__random_state: 0; NN__solver: adam	0.709