Model performance using recursive feature elimination (RFE).

(A) Area under the curve (AUC) values for the training and test dataset obtained during the PSC NCV analysis with different numbers of features selected (i.e., 10, 30, 50, and 100) during the RFE step. (B) The same results with nPSC. Note that the y-axes are on different scales. The dashed line at 0.7 indicates the threshold for feature selection. The size of the feature combination set for each independent run is shown in the box on top. (C) Train and validation AUC obtained using the reduced set of features previously selected with PSC NCV. (D) Train and validation AUC obtained using the reduced set of features previously selected with nPSC NCV. (E-F) Train and validation AUC obtained after feature size reduction based on the training feature importance. (G-H) Precision (P), recall (R) and F1 score (F1) for the AET outcomes (blue indicating AET success, i.e., eradication, and red indicating AET failure) obtained using a further feature size reduction based on the training feature importance. The dashed line at 0.8 provides a common reference. The best performing models are indicated with a black square.