Table 1.

The performances of the six machine learning models with all features (column 2) and with the active features (column 3) in terms of the AUC.

Algorithm	C-score for classification with all features	C-score for classification with active features	V-score for classification with active features
L1-regularized LR	0.931	0.921	0.897
LR w/early stopping	0.904	0.923	0.884
Random forest	0.854	0.878	0.666
Convolutional neural network	0.779	0.850	0.650
Conditional inference forest	0.801	0.822	Did not run
Multi-layer perceptron	0.695	0.489	Did not run

The V-scores for classification with the active features (column 4) indicate each model’s generalizability. We used 1000 seeds to account for the random number variance.