Precision-Recall Curve

Preview in your theme

status="ready" · three injury-risk models over one season holdout, 81 injury weeks in 1,394 player-weeks. The dashed floor at 5.8% is what flagging at random scores, so height above it — not above zero — is the skill; the faint arcs are iso-F1 guides. The rule-based curve ships no reported AUC-PR, so its legend prints the drawn area and says so. Colour is never the only channel: each curve carries its own dash pattern.

Injury risk flags · season holdout

prevalence 5.8% · Gradient boosting v4 AUC-PR 0.583 · Logistic baseline AUC-PR 0.289 · Minutes-load rule AUC-PR 0.105

Gradient boosting v4AUC-PR 0.583
Logistic baselineAUC-PR 0.289
Minutes-load ruleAUC-PR 0.105 (drawn)
no-skill baseline · prevalence 5.8%

Height above the dashed floor — not above zero — is the skill: flagging at random already scores precision 5.8% at every recall. The faint dashed arcs are iso-F1 guides: every point on one scores the same F1, so a curve that crosses a higher arc is winning even where two models intersect.

Gradient boosting v4 reports AUC-PR 0.583 but the 29 operating points drawn here enclose 0.495; Logistic baseline reports AUC-PR 0.289 but the 29 operating points drawn here enclose 0.276. The reported figure is the one shown — average precision steps between cutoffs and covers the full sweep, so straight chords over a shipped subset always read differently.

Precision-recall curves: 3 models on the unit square, recall across, precision up. Gradient boosting v4 AUC-PR 0.583, Logistic baseline AUC-PR 0.289, Minutes-load rule AUC-PR 0.105. The dashed floor at precision 5.8% is the prevalence — what flagging at random scores at every recall — so height above the floor, not above zero, is the skill. Exact values follow in the data table. Each curve lists up to 12 operating points, evenly spaced by rank.
Model	Recall	Precision	F1
Gradient boosting v4 — AUC-PR 0.583, 29 operating points	0.093	0.939	0.169
Gradient boosting v4	0.247	0.830	0.380
Gradient boosting v4	0.372	0.724	0.491
Gradient boosting v4	0.625	0.481	0.543
Gradient boosting v4	0.791	0.312	0.448
Gradient boosting v4	0.928	0.171	0.289
Gradient boosting v4	0.972	0.119	0.212
Gradient boosting v4	0.995	0.081	0.150
Gradient boosting v4	0.999	0.070	0.130
Gradient boosting v4	1.000	0.062	0.116
Gradient boosting v4	1.000	0.060	0.112
Gradient boosting v4	1.000	0.058	0.110
Logistic baseline — AUC-PR 0.289, 29 operating points	0.017	0.770	0.033
Logistic baseline	0.065	0.621	0.118
Logistic baseline	0.150	0.488	0.229
Logistic baseline	0.366	0.310	0.336
Logistic baseline	0.539	0.225	0.317
Logistic baseline	0.750	0.148	0.248
Logistic baseline	0.861	0.114	0.201
Logistic baseline	0.961	0.080	0.148
Logistic baseline	0.986	0.070	0.130
Logistic baseline	0.998	0.062	0.116
Logistic baseline	0.999	0.060	0.112
Logistic baseline	1.000	0.058	0.110
Minutes-load rule — AUC-PR 0.105, 29 operating points	0.003	0.281	0.005
Minutes-load rule	0.012	0.229	0.023
Minutes-load rule	0.034	0.193	0.057
Minutes-load rule	0.138	0.144	0.141
Minutes-load rule	0.244	0.124	0.164
Minutes-load rule	0.456	0.100	0.163
Minutes-load rule	0.625	0.086	0.152
Minutes-load rule	0.836	0.072	0.132
Minutes-load rule	0.930	0.065	0.122
Minutes-load rule	0.979	0.061	0.115
Minutes-load rule	0.994	0.059	0.112
Minutes-load rule	0.999	0.058	0.110

status="empty" · a valid contract with no operating points at all, worded so it cannot be mistaken for a failed fetch — the prevalence baseline is still known, the curves are not.

Injury risk flags · season holdout

Precision against recall, one curve per model

No operating points to plot

A precision-recall curve needs the (recall, precision) pairs an evaluation sweep emits, plus the prevalence that anchors its baseline — not a single confusion matrix and not raw scores.

status="loading" · a square skeleton holding the plot box, a decaying-precision silhouette and its dashed baseline, aria-hidden, honouring prefers-reduced-motion, with no timer involved.

Injury risk flags · season holdout

Precision against recall, one curve per model

status="error" · the fetch failed. Try again only appears when onRetry is passed, and clicking it flips the card to ready.

Injury risk flags · season holdout

Precision against recall, one curve per model

Couldn't load these curves

The data source didn't respond.

Loading preview…

This component is part of Pro

The install command, prompt and concept map are included with a Pro license — one purchase unlocks every Pro component, forever.

Unlock with Pro