False negative

Real fraud that the system missed and treated as legitimate.

A fraudster gets through, the transaction or application is approved, and nobody notices until the loss shows up later.

A fraudulent event that a detection system classifies as legitimate. It is one of two ways a fraud system can be wrong, and it is the counterpart of a false positive, which flags an honest customer as fraud.

The four possible outcomes of a fraud decision are:

• True positive, where fraud is correctly flagged.
• True negative, where a legitimate event is correctly allowed.
• False positive, where a legitimate event is wrongly flagged.
• False negative, where fraud is wrongly allowed.

Teams measure false negatives indirectly:

1. They confirm fraud cases later, through disputes, chargebacks, or investigations.
2. They check which of those cases the system had approved.
3. They calculate the detection rate, also called recall, which is the share of all fraud that the system caught. The miss rate is the share it let through.

Two features make false negatives hard to see:

• Delay, since fraud is often confirmed weeks or months after the event.
• Hidden cases, since fraud that is never reported or discovered never gets counted.

The false negative sits in a trade-off with the false positive. A stricter threshold lowers false negatives and raises false positives, and a looser one does the opposite. In biometric matching, the similar error is a false acceptance, in which the wrong person is accepted.

A false negative sits next to terms that are easy to mix up:

• False positive is the opposite error, in which an honest customer is flagged as fraud.
• False acceptance rate (FAR) is the biometric version of letting the wrong person through.
• Fraud rate is the share of events that are fraudulent, which false negatives push higher.