roc auc

**ROC Curve & AUC Score** **Overview** The ROC (Receiver Operating Characteristic) curve and AUC (Area Under the Curve) are performance metrics for binary classification problems specifically at **various threshold settings**. **The Problem with "Accuracy"** If you have 99 "Good" emails and 1 "Spam" email. A model that says "All Good" has 99% accuracy but tells you nothing. **ROC Curve** It plots: - **X-axis**: False Positive Rate (FPR) - "Crypto scams labeled as legitimate." - **Y-axis**: True Positive Rate (TPR/Recall) - "Spam correctly labeled as spam." As you lower the threshold (e.g., mark it spam if probability > 10% vs > 90%), the TPR goes up, but FPR also goes up. The curve visualizes this trade-off. **AUC (Area Under Curve)** A single number summary of the curve (0.0 to 1.0). - **0.5**: Random guessing. - **1.0**: Perfect classifier. - **0.9**: Excellent. **Interpretation** "An AUC of 0.8 means there is an 80% chance that the model will rank a random positive instance higher than a random negative instance." Use AUC when you care about *ranking* ability, not just the hard label.