When uncertainty quantification fails: structural vulnerabilities and corrections for selective prediction under clinical class imbalance
Clinical pipelines deploying selective prediction under imbalance should adopt threshold-aligned uncertainty measures and class-stratified evaluation to reduce class-specific risk imbalance, supporting safer selective prediction under the evaluated conditions without silently excluding the most vulnerable patients.
W. Endrizzi, F. Ragni, R. Marchesi et al.
· 0 citations