AI glossary
False Negative
A false negative is a case where a classification model or test predicts the negative outcome (for example, “no disease,” “not spam,” “not fraud”) when the actual, correct answer is positive. In statistics this is also called a Type II error.
False negative vs false positive
A false positive is the opposite kind of mistake: the model predicts the positive outcome when the true answer is negative (for example, flagging a legitimate email as spam). Both are counted as errors, but they are usually not equally costly; which one matters more depends entirely on the specific use case.
In a spam filter, a false positive means you lose an important email. A false negative means you still have to deal with junk mail. In cancer screening, a false negative means a patient goes home thinking they are healthy when they are not. A false positive means an unnecessary biopsy or anxiety, but the disease is caught.
Where it fits in the confusion matrix
A confusion matrix lays out four outcomes for a binary classifier: true positive, true negative, false positive, and false negative, comparing the model’s predictions against the actual labels. The false negative count is the number of actual positive cases the model labeled as negative, i.e., missed entirely.
When you look at a confusion matrix, you will see that false negatives sit in the quadrant where the actual class is positive but the predicted class is negative. This position highlights a critical failure mode: the model failed to detect the signal it was designed to find.
Why false negatives matter (recall)
Recall (also called sensitivity or the true positive rate) is calculated as true positives divided by the sum of true positives and false negatives. It measures what fraction of the actual positive cases the model successfully caught. A high false-negative count directly lowers recall: the more real positive cases a model misses, the lower its recall score, regardless of how it performs on negative cases.
Every false negative a model produces directly subtracts from its recall score, since recall only counts the positive cases the model actually caught. This metric is crucial when the cost of missing a positive case is high. For Precision and recall to be balanced, you often have to adjust the model’s threshold.
Real-world examples
A spam filter that lets an actual spam email land in the inbox (instead of flagging it) has produced a false negative. While annoying, this is often acceptable if it means legitimate mail gets through.
A medical screening test that reports “no disease” for a patient who actually has the disease is a false negative; this is a central concern in the design of and reporting of diagnostic tests. Missing a serious condition can lead to delayed treatment and worse outcomes.
A Facial recognition system that fails to match a person’s face to their own photo on file (a “false non-match”) is functionally a false negative for that recognition task, and error rates of this kind are one of the metrics independent testing of facial recognition systems reports. This can wrongly deny a legitimate user access to their own device or to a secure area.
Trading off false negatives and false positives
Almost every classifier has a decision threshold that can be tuned, and moving that threshold to catch more true positives (reducing false negatives) will typically also let through more false positives, and vice versa; you cannot generally minimize both at once.
Which mistake is more acceptable is a judgment call specific to the application: in medical screening or fraud detection, missing a true positive (a false negative) is often treated as far more costly than a false alarm (a false positive), while in a spam filter the opposite priority is common, since users are more annoyed by legitimate mail being blocked than by an occasional spam message getting through.
Understanding this trade-off is key to building effective Machine learning models. You must decide what kind of error your users can tolerate.
FAQ
What is a false negative in machine learning?
It is when a model predicts a negative class (e.g., “no”) but the actual label is positive (e.g., “yes”).
How is a false negative different from a false positive?
A false negative misses a positive case. A false positive incorrectly flags a negative case as positive.
Does a high false negative rate affect precision?
It affects recall more directly. Precision is calculated using false positives in the denominator. However, changing the threshold to reduce false negatives often increases false positives, which can lower precision.