Fig. 9

Results of post hoc-calibrated uncertainty quantification methods. The performance of the models is evaluated on the test set. CEs, BSs, and AUC values are shown across targets for selected uncertainty estimation methods and their Platt-scaled counterparts. Results are averaged over ten model repetitions, except for the deep ensemble models, for which five model repeats were computed. For each performance metric, the results of the best model are bold and underlined. All other bold results are statistically indistinguishable from the best result as reported in a t-test (\(p<\) 0.05)