F1 score

Definitions

1. The harmonic mean of precision and recall, computed as 2 x (precision x recall) / (precision + recall), used to summarise classifier performance in a single number. The harmonic mean is chosen deliberately over the arithmetic mean because it punishes imbalance: a model with perfect precision and near-zero recall scores near zero rather than around fifty percent. F1 is the standard metric on imbalanced datasets, where raw accuracy is meaningless because always predicting the majority class already scores well. Its main limitation is that it ignores true negatives entirely and weights precision and recall equally, which rarely matches real costs; the general F-beta score lets you weight recall higher when missing a positive is expensive, as in disease screening, or precision higher when false alarms are costly.

   In plain English: A single number combining how precise and how thorough a model is. It is a better measure than accuracy when the thing you are looking for is rare.

   Example: The model's accuracy was 97%, but the F1 score for the minority class was only 0.41, exposing how poorly it detected actual churning customers.

Related Terms

• Perplexity
• Benchmark
• benchmark contamination
• confusion matrix
• precision (ML)
• recall (ML)
• K-Nearest Neighbors