False Negative
What is a false negative?
What does a false negative mean in identity verification?
- 1:1 biometric comparison. A comparison between samples from the same person that incorrectly returns a non-match is called a false non-match. False non-match rate (FNMR) measures how often genuine comparisons produce this error. The opposite error is a false match, in which samples from different people incorrectly match.
- 1 biometric identification. A false negative identification occurs when a search fails to return the enrolled person as a candidate above the selected threshold. False negative identification rate (FNIR) measures this error.
- Presentation attack detection (PAD). A missed attack may be described generally as a false negative, although PAD standards use specific metrics for attacks accepted as bona fide presentations and genuine presentations classified as attacks.
- Sanctions or watchlist screening. A false negative occurs when the system fails to return a genuine match to a listed person or entity.
What causes false negatives?
Why are false negatives important?
How can Regula support false-negative analysis?
FAQ
Is a legitimate customer flagged as fraud a false negative?
No. In a system that treats fraud as the positive class, flagging a legitimate customer as fraudulent is a false positive. A false negative occurs when the system fails to flag genuine fraud. If a face matcher fails to match a legitimate customer with their own reference image, the specific biometric term is false non-match. The errors come from different checks and require different metrics.
How is the false-negative rate calculated?
The false-negative rate is the number of false negatives divided by all ground-truth positive cases: False-negative rate = false negatives ÷ (true positives + false negatives). If fraud is the positive class, the denominator contains all confirmed fraudulent cases in the test set. Biometric comparison and identification use specific metrics such as FNMR and FNIR.
Is every failed identity verification a false negative?
No. A legitimate applicant may fail because the image could not be captured, a required document was missing, a business rule stopped the process, or a biometric comparison returned a false non-match. Only the last example is a biometric comparison error. End-to-end rejection rates and component error rates should be reported separately.
Does changing the threshold reduce false negatives?
It can, although the effect depends on how the score is defined. For face similarity scores, lowering the match threshold generally reduces false non-matches and increases false matches. Fraud systems may assign scores in the opposite direction, so teams should confirm what higher and lower scores mean before changing a threshold.
.webp)