Formula Used
Sensitivity = TP ÷ (TP + FN)
Specificity = TN ÷ (TN + FP)
Precision = TP ÷ (TP + FP)
F1 Score = 2 × Precision × Recall ÷ (Precision + Recall)
How to Use
- Select a calculation mode.
- Enter confusion counts, labels, probabilities, or model rows.
- Choose confidence and precision settings.
- Add false-negative and false-positive costs when needed.
- Calculate, review the interpretation, then export your report.
Example Data
| Scenario | TP | FN | FP | TN | Sensitivity |
|---|---|---|---|---|---|
| Medical screening | 95 | 5 | 18 | 82 | 95.00% |
| Fraud detection | 76 | 24 | 12 | 188 | 76.00% |
| Defect detection | 88 | 12 | 9 | 91 | 88.00% |
Frequently Asked Questions
What is sensitivity in machine learning?
Sensitivity measures correctly detected positive cases. It is also called recall. Higher values mean fewer positives are missed.
How is sensitivity different from precision?
Sensitivity focuses on actual positives. Precision focuses on predicted positives. Both metrics answer different model-quality questions.
How is sensitivity different from specificity?
Sensitivity measures positive detection. Specificity measures negative rejection. Strong classifiers often balance both values.
Can sensitivity equal one hundred percent?
Yes, when every positive case is detected. False negatives must equal zero. Specificity may still be weak.
What happens when no positive cases exist?
The denominator becomes zero. Sensitivity is then undefined. Add positive examples before interpreting recall.
Which threshold gives the best sensitivity?
Lower thresholds usually increase sensitivity. They may also increase false positives. Choose thresholds using real costs.
What is macro-average sensitivity?
Macro averaging treats every class equally. It ignores class frequency. This helps reveal minority-class performance.
What is weighted-average sensitivity?
Weighted averaging uses class support. Larger classes influence the result more. It reflects overall dataset composition.
Why use a confidence interval?
A confidence interval shows estimation uncertainty. Wider intervals suggest limited sample evidence. Larger samples usually narrow intervals.