False Positive
What is a false positive?
Which metrics describe false positives in identity verification?
- 1:1 biometric comparison. A comparison between samples from different people that incorrectly returns a match is called a false match. False match rate (FMR) measures how often impostor comparisons produce this error. The opposite error is a false non-match, in which samples from the same person fail to match.
- 1 biometric identification. A false positive identification occurs when a search involving a person who is not enrolled returns one or more candidates above the selected threshold. False positive identification rate (FPIR) measures this error.
- Document validation. When a genuine document is incorrectly classified as fraudulent, document-specific testing may report a document false rejection. This is different from accepting a fraudulent document, which is a document false acceptance.
- Presentation attack detection (PAD). A bona fide presentation classified as an attack may be described generally as a false positive. PAD standards use their own error metrics, which should be named in technical reports.
- Sanctions or watchlist screening. A false positive occurs when a person or entity that is not the listed party is incorrectly returned as a potential match. This result usually requires review rather than automatic rejection.
What causes false positives?
Why are false positives important?
How can Regula support false-positive analysis?
FAQ
How is a false positive different from a false negative?
A false positive reports a target that is not present; a false negative misses a target that is present. If fraud is the target, a legitimate customer incorrectly flagged as fraudulent is a false positive, while undetected fraud is a false negative. The positive class must be stated before either label can be interpreted correctly.
Can a false positive allow a fraudster to pass face matching?
The general phrase can be misleading here. In fraud detection, a fraudster who passes is a false negative because the fraud was missed. In 1:1 face matching, an impostor who incorrectly matches another person’s reference produces a false match, sometimes loosely called a biometric false positive. Reports should use false match and FMR for the biometric comparison so the two errors are not confused.
Is every rejected legitimate customer a false positive?
No. The rejection is a false positive only if a system incorrectly classified the customer or their evidence as the target it was meant to detect, such as fraud. A customer may also fail because required evidence is missing, an image cannot be processed, or a business rule prohibits the transaction. These events should be measured separately.
How is the false-positive rate calculated?
The false-positive rate is the number of false positives divided by all ground-truth negative cases: False-positive rate = false positives ÷ (true negatives + false positives). If fraud is the positive class, the denominator contains all confirmed legitimate cases in the test set. Biometric comparison and identification use specific metrics such as FMR and FPIR.
Does changing the threshold reduce false positives?
It can, but the direction depends on the score. Raising a face-similarity threshold generally reduces false matches and increases false non-matches. For a fraud-risk score, changing the alert threshold may have the reverse numerical direction. Any threshold change should be assessed against both error types using representative data.
.webp)