Figure 3.

Workflow for model stacking with nested CV. For each training data set in the outer CV loop (dark blue bars on top) complete with true resistance status of samples (red bars), an inner CV loop is run (light blue bars). The full set of predictions (yellow bars) obtained from the test sets of the inner CV are used to train a stacking model to ideally combine predictions from each of the components. At the same time, full component models are trained on the training data set (blue bars within component models). Subsequently, predictions are made by all full component models on the test dataset (green bars on top). Predictions are made by the stacking model using the component model predictions as input features. Finally, performance metrics are obtained by scoring predictions of each model type against the true resistance status of test set samples.