Fig. 2
From: Susceptibility of AutoML mortality prediction algorithms to model drift caused by the COVID pandemic
ROC- curves (top left), PR-curves (top right) and Venn-Diagram (bottom) of the best pre-pandemic model of each family. The dashed line in the precision-recall plot showed the baseline which is the precision that would be achieved if the model always predicted the negative class. The Venn Diagram includes the 100 most important features of the best model of each family. Intersections without numbers are empty sets, the outer areas (84,14,41,47,15, and 14 features respectively) occur only in the top 100 of the respective model. The centre of the diagram shows the 8 parameters that occur in each of the models. GLM: Generalized Linear Model, DRF: Default Random Forest, GBM: Gradient Boosting Machine, XGB: eXtreme Gradient Boosting, RPC: red packed cells, ASA: American Society of Anesthesiologists, INR: International Normalized Ratio