Use this tool when the positive class is rare and you care about how many of your positive calls are correct, for example fraud, disease screening or defect detection. It shows the trade-off between precision and recall across all cut-offs.
About the precision-recall curve
A precision-recall curve plots precision (the share of predicted positives that are truly positive) against recall (the share of true positives that are found) for every cut-off of a continuous score. Average precision summarises the curve in one number.
When to use it
Prefer it to the ROC curve for imbalanced data, where a high AUC can hide many false positives among a small number of true cases.
What this tool reports
The precision-recall curve with the no-skill baseline.
Average precision and the precision and recall at the chosen cut-off.
Tables you can download as CSV and a ready-to-use APA7 results paragraph.
An editable figure: colours, text, sizes and line widths.
Assumptions to check
The outcome has two classes, with the class of interest coded as positive.
Cases are independent.
There are enough positive cases for the curve to be stable.
How to report it in APA7
Report the sample size and the share of positive cases, the average precision, and the precision and recall at the chosen cut-off, and say how the cut-off was chosen. The tool drafts this paragraph for you.
Frequently asked questions
Is the Precision-Recall Curve tool free?
Yes. You can run the analysis and read the results without an account. Create a free account or log in only when you want to download the results.
When should I use it instead of a ROC curve?
When positives are rare. Precision reacts to false positives in a way the ROC curve does not.
What is the baseline?
A model with no skill has a precision equal to the share of positive cases in the data.
Is my data uploaded to a server?
The analysis runs in your browser. Your data stays on your device while you work with the tool.