Fig. 2
From: Synergy conformal prediction applied to large-scale bioactivity datasets and in federated learning
Top panels: efficiency for the active class for Set 1 using the different conformal predictors at a range of significance levels (0.1–0.3). Results for RDKit descriptors left and fingerprints right. Bottom panels: pairwise comparison (Wilcoxon signed-rank test with Bonferroni correction for multiple testing, 0.05 significance level, across all significance levels and datasets) of methods on rows with methods on columns, significantly better result is indicated in blue, significantly worse result in red. p-values are indicated in the figure. For example, in the bottom left panel we can see that SCP 10 is significantly worse than all other methods it is compared with