Classification threshold and ROC table

Inspect precision, recall, false-positive rates and ROC AUC at each distinct score threshold.

The example value has been entered. Check it immediately or change the value.

Input is processed in the browser. Refreshing returns a sample value.

Result

Instructions and example

score≥threshold; AUC=ΣΔFPR·(TPR₁+TPR₂)/2

The default 4 observations give ROC AUC 0.75. Threshold 0.7 predicts 3 positives with recall 1.

At most 200 rows, scores from 0 to 1, and both actual classes are required. Tied scores move together; no threshold is automatically recommended. Floating-point results display at most 12 significant digits.

Frequently asked questions

Are scores equal to the threshold positive?

Yes, score≥threshold. The +∞ row is the starting point with no predicted positives.

Does a high ROC AUC prove probability calibration?

No. AUC describes ranking separation, not probability calibration or real error costs.

Calculation criteria and reference materials

Verification of calculation criteria: · Calculation·verification principles · Report errors