Fig. 3

Model performances on the unseen benchmark dataset. A Balanced accuracy for all machine learning algorithms. B Precision-Recall curve based on all variables and the corresponding area under the precision recall curve (AUCPR). C Receiver operating characteristics (ROC) curve based on all variables and the corresponding area under the curve (AUC). D Balanced accuracy, based on the number of most important features selected, the dashed grey line illustrates the selected threshold for LR and SVM, dashed green line for Xgboost. Confusion matrices for E Xgboost based on all variables, F Xgboost based on top 45 variables, G LR based on all variables, H LR with the top 75 important variables, I SVM based on all variables and J SVM based on top 75 most important variables. Numbers in brackets in (B-D) correspond to the 95% confidence intervals determined by 500 bootstrappings. Logistic regression (LR), support vector machine (SVM), naive bayes (NB), light gradient boosting machine (LightGBM), multilayer perceptron (MLP), k-nearest neighbors (KNN) and random forest (RF)