An Introduction to ROC Analysis.¶
Fawcett, T. (2006). An Introduction to ROC Analysis. Pattern Recognition Letters, 27(8), 861-874.
Cited by¶
6 citations across 6 artifacts.
Each citation links to the sentence it supports in the citing article.
Primes¶
- Bycatch
- The precision/recall trade-off and ROC vocabulary inherit directly into fisheries-selectivity design, where gear ROC curves trade target retention against non-target retention.
This sourceStandard reference on ROC curves, precision/recall, and the false-positive/true-positive trade-off including low-base-rate effects.
- The precision/recall trade-off and ROC vocabulary inherit directly into fisheries-selectivity design, where gear ROC curves trade target retention against non-target retention.
- Signal Detection Theory
- Second, a machine-learning fraud classifier outputs a probability score; the evidence distributions are the score densities for legitimate versus fraudulent transactions, their separation summarized by AUC (sensitivity), and the decision threshold on the score is the criterion.
This sourceStandard reference on ROC curves, AUC, and decision-threshold tuning for machine-learning classifiers.
- Second, a machine-learning fraud classifier outputs a probability score; the evidence distributions are the score densities for legitimate versus fraudulent transactions, their separation summarized by AUC (sensitivity), and the decision threshold on the score is the criterion.
Domain-specific¶
- Binary classification
- Label Shift
- Each cell of the factorisation carries a different correction — feature-space reweighting, relearning, and prior reweighting respectively — and applying one cell's remedy to another's fault wastes effort or fails outright. Not a discrimination failure. Because P(X|Y) is held fixed, the model's learned ranking is provably preserved: ROC AUC is prior-invariant and does not move when only P(Y) shifts
This sourceThat ROC curves and AUC are insensitive to the class prior, so discrimination is unchanged when only P(Y) moves.
- Each cell of the factorisation carries a different correction — feature-space reweighting, relearning, and prior reweighting respectively — and applying one cell's remedy to another's fault wastes effort or fails outright. Not a discrimination failure. Because P(X|Y) is held fixed, the model's learned ranking is provably preserved: ROC AUC is prior-invariant and does not move when only P(Y) shifts
Mechanisms¶
- Decision Threshold Rule
- There is no free setting: every position trades one error against the other, a tradeoff the lender reads off a sensitivity-versus-specificity curve.
This sourceExplains that varying a decision threshold moves a classifier through a tradeoff between true-positive and false-positive rates represented in ROC space.
- There is no free setting: every position trades one error against the other, a tradeoff the lender reads off a sensitivity-versus-specificity curve.
- Receiver Operating Characteristic Review
- It also exposes a deeper truth — if the curve hugs the diagonal, no threshold will satisfy both, and the fix is a better matcher, not a cleverer cutoff.
This sourceShows that a ROC curve on the diagonal has random discrimination, which threshold changes cannot repair; better performance requires a better classifier.
- It also exposes a deeper truth — if the curve hugs the diagonal, no threshold will satisfy both, and the fix is a better matcher, not a cleverer cutoff.
Verification¶
Does it exist? Confirmed. This work's DOI resolves to a registered record, which fixes its identity. That is all it fixes.
Does it back the claim? Not recorded. Neither this nor any other of the 6 citations of this work carries a recorded support check.
Support is checked per citation rather than per work — the same source can be cited soundly in one article and wrongly in another. Per-citation recording began recently, so a citation with no recorded check is a gap in the record rather than evidence it went unchecked.
See how references were verified.
Registry ID ref:bc55cfe19d4d · see in the full table