Landmark 4 stops to get here · leads to 2

False Positive

A case the model labels positive that is actually negative, such as a legitimate email sent to spam. Statisticians call it a Type I error.

Your route here

4 stops · basics first
  1. Machine Learning ✓ understood

    Building systems that learn patterns from data instead of following hand-written rules, getting better at a task as they see more examples.

  2. Supervised Learning ✓ understood

    Learning from examples paired with the correct answer, so a model can predict answers for new inputs it hasn't seen.

  3. Classification ✓ understood

    A supervised learning task where the model assigns each input to one of a fixed set of categories, such as spam or not spam.

  4. Confusion Matrix ✓ understood

    A table showing true positives, true negatives, false positives, and false negatives for classification evaluation.

  5. False Positive · you are here ✓ understood

Picture it

PREDICTEDpositivenegativeACTUALpositivenegativeTP42FN6FP8TN44precision = TP / (TP + FP)= 42 / 50 = 0.84recall = TP / (TP + FN)= 42 / 48 = 0.88
The FP cell counts negatives the model wrongly called positive; notice it appears in precision = TP/(TP+FP), pulling it down.

A false positive is a false alarm: the model says “yes” when the true answer is “no.” A spam filter that sends your colleague’s email to the junk folder has produced one. So has a fraud system that blocks a legitimate purchase, or a screening test that flags a healthy patient.

The name comes from the two halves. “Positive” is what the model predicted; “false” means the prediction was wrong. In statistics the same mistake is called a Type I error, rejecting a null hypothesis that was actually true.

Where it sits

In a confusion matrix, false positives are the actual negatives that landed in the predicted-positive column. The diagram puts them next to the true positives, and that pairing is what precision measures: TP / (TP + FP). Every false positive drags precision down.

Two other metrics are built from the same cell. The false positive rate, FP / (FP + TN), is the share of all real negatives that got flagged; it’s the x-axis of an ROC curve. Specificity is its complement, 1 minus the FPR.

Trading one error for another

Most classifiers output a score, and a threshold turns it into yes or no. Raise the threshold and the model flags fewer cases, so false positives fall, but more real positives slip through as false negatives. Lower it and the reverse happens. You can’t drive both to zero with the same model, so you pick the balance based on which mistake costs more. A spam filter should rarely bury real mail. A cancer screen can tolerate false alarms, because a follow-up test is cheaper than a missed diagnosis.

The catch

When positives are rare, false positives can outnumber true ones even for a model that looks accurate. Take 10,000 transactions, 10 of them fraudulent. A model that catches all 10 and wrongly flags just 1% of the 9,990 honest ones produces about 100 false alarms. Its accuracy is about 99%, yet only about 1 in 11 flagged transactions is really fraud. That’s why accuracy alone misleads on imbalanced data, and why teams report precision and the false positive rate alongside it.

Where it sits

Before this

Confusion Matrix
False Positive

Explore nearby

Sources

  1. Google, Machine Learning Crash Course: "Classification: Accuracy, recall, precision, and related metrics" . Google for Developers
  2. Christopher D. Manning, Prabhakar Raghavan and Hinrich Schütze, Introduction to Information Retrieval . Cambridge University Press, 2008; Evaluation of unranked retrieval sets