Which classification metrics should you use (accuracy, precision, recall, F1, AUC) and why?

Metric choice depends on the business cost of errors. - Accuracy: misleading on imbalanced data - Precision: minimize false positives - Recall: minimize false negatives - F1: balance precision and recall - ROC-AUC/PR-AUC: ranking quality; PR-AUC is better for imbalance Always tie metrics to business impact and choose thresholds explicitly.