Precision and recall pull in opposite directions. Tighten a classifier to reduce false positives and you miss more true positives. Loosen it and the reverse happens. The F1 score combines both into a single number, the harmonic mean of precision and recall.

The harmonic mean punishes imbalance. If either precision or recall is low, F1 drops sharply. A model with 90 percent precision and 10 percent recall scores around 18, not 50. That behaviour makes F1 useful when both errors matter.

When F1 helps

• Imbalanced classes
• Tasks where both false positives and false negatives cost
• Comparing models on a single metric
• Reporting on information retrieval systems

F1 is not always the right choice. When one error type is far worse, weighted variants or a custom cost function serve better. F1 assumes precision and recall deserve equal weight, which is a modeling decision, not a law.