ROC-AUC Explained
ROC (Receiver Operating Characteristic)
ROC is a curve/graph that shows how well your model separates classes.
- X-axis → False Positive Rate (FPR)
- Y-axis → True Positive Rate (TPR)
It plots performance at different probability thresholds.
AUC (Area Under Curve)
AUC represents the area under the ROC curve. Its value ranges from 0 → 1.
| AUC Score | Meaning |
|---|---|
| 1.0 | Perfect model |
| 0.9 – 0.99 | Excellent |
| 0.8 – 0.89 | Good |
| 0.7 – 0.79 | Okay |
| 0.5 | Random guessing |
| < 0.5 | Worse than random |
ROC-AUC measures:
How well your model ranks positive vs negative samples
Higher AUC → better separation. Lower AUC → more confusion.
Example
Suppose:
- Cancer = Positive
- Healthy = Negative
If the model gives higher probabilities to Cancer and lower to Healthy → AUC will be high (~0.9+).
Formula
- TPR (Recall) = TP / (TP + FN)
- FPR = FP / (FP + TN)
ROC plots TPR vs FPR.
Why ROC-AUC is used
- Works well for imbalanced data
- Does not depend on a fixed threshold
- Measures ranking quality