Figure 1.

Feature selection (performed separately for the two endpoints). The process uses the imputed data of 17 features of the entire patient population. Four feature ranking methods are performed (100 iterations averaged, not shown on the figure), while a fifth ranking order “Average” is obtained from averaging the results from the four methods. Groups created using the top 4–7 features of each ranking are assessed using ML models. The groups are then ranked based on the average AUC of the top 10% models trained on them. Finally, the top-ranking feature group is selected. KNN: K-nearest neighbors; MLP: multilayer perceptron; RF: random forest; XGB: extreme gradient boosting; PI: permutation importance; RFE: recursive feature elimination.