False Negative

Any detection tool can be wrong in two ways. It can raise an alert where there is no problem (a false positive) or stay silent where there is one (a false negative). The second error is usually the costly one, because nobody looks further. Examples are a malware scanner passing an infected file, a screening system missing a sanctioned name with a spelling variation, a fraud model approving a stolen card, or a medical test missing a disease. Teams estimate the share of real cases that are missed, often called the miss rate, by testing on known cases. Lowering a threshold to catch more usually raises false positives and workload. The balance depends on the harm each error causes. Auditors and model validators hunt for these misses first, since each one is exposure the organisation does not know it carries.

An online payments firm flags card transactions above a set amount. Auditors test the rule with 50 known fraud patterns and find that 10 of them, spread across small purchases, were never flagged. Those ten are false negatives, and the firm adds a rule that totals spending across days.