False Negative
When a model makes a mistake, the kind of mistake matters. Calling a spam email “not spam” is a very different failure than calling a normal email “spam,” even though both are just “wrong.”
What a false negative means
A false negative happens in a binary classification task when the true label is positive, but the model predicts negative. In plain terms: the thing you were trying to catch is really there, and the model misses it. In a confusion matrix, it’s the count in the “actual positive, predicted negative” cell. This is tightly connected to recall (also called sensitivity): recall drops when false negatives rise, because recall measures how many real positives you successfully found.
Why it matters in real systems
False negatives are costly when missing a positive case is dangerous or expensive. Examples:
- Disease screening: a patient has the disease, but the model predicts “healthy.”
- Fraud detection: a fraudulent transaction is labeled “legitimate,” leading to losses.
- Churn prediction: a customer who will leave is predicted to stay, so you don’t intervene.
In these settings, teams deliberately tune the decision threshold to reduce false negatives, even if it increases false positives.
How you control and measure it
Many models output a probability (e.g., scikit-learn’s LogisticRegression via predict_proba). If you lower the threshold for predicting “positive,” you typically catch more positives (fewer false negatives) but trigger more false alarms. Metrics that directly reflect false negatives include recall and the false negative rate (FNR = FN / (FN + TP)).
A false negative occurs when a classifier predicts the negative class for an instance that is truly positive (e.g., labeling a fraudulent transaction as legitimate). It is a key cell in the confusion matrix and directly drives metrics such as recall/sensitivity and the false negative rate. Minimizing false negatives is critical in high-miss-cost domains like medical screening, safety monitoring, and fraud detection.
Imagine a smoke alarm that stays silent even though there’s a real fire. That’s a false negative: something bad is present, but the system says “no problem.”
In supervised learning, a false negative happens when an AI model misses a case it should have caught. For example, a medical test model might label a sick patient as healthy, or a fraud detector might let a fraudulent transaction pass as normal. False negatives matter because they can feel reassuring in the moment, but they can lead to serious consequences later—missed treatment, undetected fraud, or unsafe content slipping through.