A hospital readmission model will trigger costly interventions. The team wants the predicted probability itself to be trustworthy because downstream decisions depend on the confidence score, not just the positive or negative label. Which metric should be emphasized?
Answer & explanation
Correct answer: B
WHY B: When calibrated probability quality matters, log loss is a strong choice because it scores the predicted probabilities directly. WHY NOT A: F1 depends on a threshold and ignores most probability information. WHY NOT C: ROC AUC measures ranking, not calibration quality. WHY NOT D: Balanced accuracy does not evaluate probabilities. WHY NOT E: MAPE is a regression metric and not standard for probability classification.