Calculation Results
Enter data to calculate the area.
Step-by-step calculation
No calculation yet.
Threshold analysis
| Threshold | TP | FP | TN | FN | TPR | FPR | Precision | Recall | Specificity | F1 | Youden J |
|---|
Calculator Inputs
Calculation and processing options
Export Results
Export calculated metrics, curve points, thresholds, and model comparisons.
Formula Used
ROC AUC measures ranking performance across every classification threshold. It plots true positive rate against false positive rate. Larger values usually indicate stronger discrimination.
TPR = TP / (TP + FN) FPR = FP / (FP + TN) Specificity = TN / (TN + FP) Precision = TP / (TP + FP) Recall = TP / (TP + FN) F1 = 2 × Precision × Recall / (Precision + Recall) Youden J = Sensitivity + Specificity − 1 Trapezoidal area = Σ [(xᵢ₊₁ − xᵢ)(yᵢ₊₁ + yᵢ) / 2]
Precision–recall AUC focuses on positive-class retrieval. It is especially informative for imbalanced datasets. Average precision uses a stepwise recall-weighted precision sum.
How to Use
- Select binary, custom, multiclass, or comparison mode.
- Enter labels, scores, probabilities, or XY coordinates.
- Choose the primary metric and integration method.
- Set partial-area limits and threshold criteria.
- Enable data processing and confidence interval options.
- Press Calculate AUC to generate the report.
- Inspect charts, thresholds, metrics, and calculation steps.
- Export the report as CSV, JSON, PDF, PNG, or SVG.
Example Data
| Actual label | Predicted score | Expected role |
|---|---|---|
| 1 | 0.95 | High-confidence positive |
| 0 | 0.72 | Possible false positive |
| 1 | 0.67 | Moderate positive |
| 0 | 0.40 | Moderate negative |
| 1 | 0.35 | Possible false negative |
| 0 | 0.10 | High-confidence negative |
Frequently Asked Questions
What does ROC AUC measure?
ROC AUC measures how well scores rank positive cases above negative cases. It considers every possible threshold. A value near one indicates strong ranking.
Is AUC the same as accuracy?
No. Accuracy uses one selected classification threshold. AUC evaluates performance across many thresholds.
When should I use PR AUC?
PR AUC is useful when positive cases are rare. It focuses on precision and recall. This avoids an overly optimistic specificity effect.
What is average precision?
Average precision summarises a precision–recall curve using stepwise recall changes. It may differ from trapezoidal PR AUC. Both should be labelled clearly.
What is partial AUC?
Partial AUC measures only a selected X-axis interval. It helps evaluate clinically or operationally relevant ranges. Normalisation supports comparisons across interval widths.
What does AUC below 0.5 mean?
It may indicate reversed score direction. Swapping positive and negative ranking can improve interpretation. Data errors should also be checked.
Can two models share the same AUC?
Yes. Their curves may cross at different thresholds. Operational performance can still differ substantially.
How is the best threshold selected?
This calculator supports Youden’s J, maximum F1, and nearest top-left distance. The correct choice depends on error costs. Domain requirements should guide deployment.
Why use bootstrap confidence intervals?
Bootstrap intervals estimate uncertainty by resampling observations. Wider intervals indicate less stable estimates. More samples usually improve interval precision.
How is multiclass AUC calculated?
Each class can be compared against all remaining classes. Per-class values are then macro, weighted, or micro averaged. One-vs-one summaries are also available.
Does curve smoothing change AUC?
This calculator smooths only the displayed chart. Calculations use the original processed points. Reported metrics therefore remain reproducible.