F1 score

The F1 score combines precision and recall into a single number that is only high when both are high.

It works like rating a restaurant on food and hygiene. A place with brilliant food and a filthy kitchen should not average out to "pretty good" — one bad half should drag the whole rating down. F1 behaves that way on purpose.

Mathematically, F1 is the harmonic mean: F1 = 2 × (precision × recall) ÷ (precision + recall). A harmonic mean punishes imbalance. Precision of 1.0 with recall of 0.1 gives an F1 of 0.18, not the 0.55 a plain average would show.

precision   recall   plain average    F1
   0.9        0.9        0.90        0.90
   1.0        0.1        0.55        0.18   ← the useless model is exposed

Use F1 when the classes are imbalanced and you need one number to compare models. Its weakness: it hides which half is failing, and it ignores true negatives entirely. Look at precision and recall separately before trusting it.

Where to go next#

• Full lesson: Model evaluation
• Related terms: precision, recall, accuracy, roc-auc