← Visualization gallery
Prediction · Reporting

Precision–Recall Curve

Plots precision (PPV) against recall (sensitivity) across thresholds — the discrimination figure to use when the outcome is rare and ROC/AUC would look deceptively strong.

Precision–Recall Curve: Plots precision (PPV) against recall (sensitivity) across thresholds — the discrimination figure to use when the outcome is rare and ROC/AUC would look deceptively strong.
Download SVG Open code and full worked example
When to use it

For imbalanced outcomes (rare events), where a high AUC can coexist with poor positive predictive value. The PR curve and average precision focus on performance among predicted positives, which is what matters when prevalence is low.

How to read it

The curve trades precision for recall as the threshold drops; the flat line is the prevalence (a no-skill classifier). Area under the curve (average precision) above the prevalence baseline indicates useful ranking among the positive class.

Worked example

A model score for 4,000 patients with an 18% event rate is thresholded across its range; precision and recall are computed at each threshold and plotted, against the 0.18 prevalence baseline.

Prevalence 18%; score = 0.55·event + Normal(0, 0.3) + 0.2, clipped to [0,1].

Result: The PR curve sits well above the 0.18 baseline with average precision ≈ 0.55 — so the model ranks true positives far better than chance even though the outcome is rare, a conclusion the ROC curve (AUC) would express less honestly at this prevalence.

Produced by
ROC, AUC & discriminationPrediction-model validation & recalibration

Reference: Gatto NM, Wang SV, Murk W, et al. Visualizations throughout pharmacoepidemiology study planning, implementation, and reporting. Pharmacoepidemiol Drug Saf. 2022;31(11):1140-1152.