False negative rate (FNR) is the proportion of actual positive cases that a model fails to detect, classifying them as negative when they are not.
This machine learning metric is calculated as the number of false negatives divided by the total number of actual positives. A model that misses 80 out of 1,000 real positive cases has a false negative rate of 8 percent.
FNR sits within the broader family of classification metrics and is closely tied to its counterpart, the false positive rate.
An overfit model can appear to have a low FNR on training data but perform significantly worse on new, unseen data, making it an unreliable measure of real-world performance.
Achieving a zero FNR is theoretically possible but difficult to achieve. In most cases, teams accept some level of false negatives and tune the model's threshold based on what the specific use case can tolerate.