Precision-Recall Curve by Class Calculator

Analysis results

Rank	Class	Support	Prevalence	AP	PR AUC	Best F1	Best threshold	Vs macro AP	Status

Class filter

Minimum threshold

Maximum threshold

Class	Threshold	Precision	Recall	F1	Specificity	TP	FP	FN	TN	Predicted positive	Cost

Add optional Model B data to compare models.

Data and class setup

Class names

Use commas. Names must be unique.

Text delimiter

Label mode

Use a pipe between multilabel actual classes.

Example preset

Paste CSV, TSV, pipe text, or JSON

Choose CSV, TSV, TXT, or JSON

Headers may be actual, class names, and weight.

Enable Model B comparison

Model B data

Actual labels and class names should match Model A.

Calculation options

Classes to analyse

Hold Ctrl or Command to select several.

Threshold selection

Custom threshold: 0.50

Minimum recall

Minimum precision

Threshold table step

False-positive cost

False-negative cost

Decimal places

Normalize probability rows

Use sample weights

Ignore invalid rows

Show row weights

Chart options

Chart title

Horizontal-axis title

Vertical-axis title

Legend position

Curve rendering

Chart height

Line width

Marker size

Annotate every N points

Class display styles

Run and export

Formula used

Precision

Precision measures correct positive predictions. It penalizes false positive decisions.

Precision = TP / (TP + FP)

Recall

Recall measures detected positive examples. It penalizes missed positive observations.

Recall = TP / (TP + FN)

F1 score

F1 balances precision and recall. It uses their harmonic mean.

F1 = 2 × Precision × Recall / (Precision + Recall)

Average Precision: the weighted sum of precision values across recall increases.

PR AUC: trapezoidal integration of precision over recall.

One-versus-rest: each selected class becomes positive. Other classes become negative.

How to use

Enter unique class names.
Add actual labels and class probabilities.
Choose classes, thresholds, costs, and chart settings.
Calculate the curves and review class summaries.
Export tables, images, data, or a report.

Example data format

actual	Cat	Dog	Bird	weight
Cat	0.82	0.12	0.06	1
Dog	0.18	0.70	0.12	1
Bird	0.09	0.21	0.70	1

Frequently asked questions

What does each curve represent?: Each curve treats one class as positive and others as negative.
Why use precision-recall curves?: They are especially useful when positive classes are rare or imbalanced.
What is Average Precision?: Average Precision summarizes precision across increasing recall levels.
Is PR AUC identical to Average Precision?: No. This calculator reports both integration methods separately.
What is the no-skill baseline?: It equals the positive prevalence for the evaluated class.
What does micro-average mean?: It pools every one-versus-rest decision before computing one curve.
What does macro-average mean?: It gives equal influence to every selected class.
What does weighted-average mean?: It weights class performance using each class support.
What does samples-average mean?: It averages label-ranking precision separately across observations.
Can I use multilabel targets?: Yes. Select multilabel mode and separate actual labels with pipes.
Can I compare two models?: Yes. Enable Model B and paste matching prediction data.