Table 3.

Performance comparison.

Performance metric	Full feature set	Reduced feature set
ROC AUC	0.863 [0.838, 0.885]	0.861 [0.838, 0.883]
PR AUC	0.443 [0.405, 0.489]	0.434 [0.414, 0.485]
Brier score	0.081 [0.050, 0.095]	0.082 [0.050, 0.095]
Log loss	0.265 [0.177, 0.305]	0.269 [0.178, 0.306]
Sensitivity	0.828 [0.805, 0.855]	0.829 [0.805, 0.852]
Specificity	0.749 [0.702, 0.786]	0.754 [0.713, 0.785]
F1-score	0.439 [0.374, 0.475]	0.439 [0.368, 0.478]

Different test set performance metric distributions across splits for the models before and after feature reduction based on feature importance. Numbers are reported as median and interquartile range.