Fig. 3
From: Susceptibility of AutoML mortality prediction algorithms to model drift caused by the COVID pandemic
AUROC and AUPR pre- and in-pandemic. Two points that belong together are each connected by a line. The respective left point represents model performance on the pre-pandemic test set, the right point on the in-pandemic test set. Changes in area under receiver-operating characteristic (AUROC) curve are depicted in the left column, changes in area under precision-recall (AUPR) curve are depicted in the right column. This graphic shows all the created models of each family. Most models suffer from a performance loss in the pandemic, not only the native ones, but also those modified by means of the “weight”, “6 months” and “scaled” methods. GLM: Generalised Linear Model, DRF: Default Random Forest, GBM: Gradient Boosting Machine, XGB: eXtreme Gradient Boosting, DL: Deep Learning, Stacked: Stacked Ensemble