##  [F1 Score](/f1-score-0) 

 Definition

A summary statistic equal to the harmonic mean of precision and recall: F1 = 2 · (precision · recall) / (precision + recall); it combines both measures into a single value that penalizes extreme imbalance between them.

 

 

 

 

 

 





## Principle

Principle

F1 balances precision and recall by giving equal weight to both; it rises only when improvements occur in precision, recall, or both, and falls when one improves at the expense of the other disproportionately.

 

 

 

 

 





## Demonstration

Demonstration

Illustrative scenario: a system with precision 0.75 and recall 0.60 yields F1 = 2·(0.75·0.60)/(0.75+0.60) ≈ 0.67; recognition: the F1 summarizes retrieval effectiveness under binary relevance at the chosen cutoff; action: analysts examine whether to prioritize precision or recall to raise F1.

 

 

 

 

## Misapplication

Misapplication

Using F1 as the sole criterion when precision and recall have different operational priorities or when relevance is graded, time-sensitive, or ranking position matters; the semantic error is collapsing multidimensional performance into a single undifferentiated score.

 

 

 

 

 





## Consequence

Consequence

F1 provides a compact comparable metric for binary relevance tasks and model selection when equal weight is appropriate; overreliance can obscure important differences in precision versus recall trade-offs or in rank-sensitive utility.

 

 

 

 

## Reversal

Reversal

When task stakeholders value recall more than precision (or vice versa), a weighted Fβ measure (β≠1) or alternative metrics that account for rank or graded relevance are more appropriate; F1's equal-weight assumption fails to reflect asymmetric costs.

 

 

 

 

 





## Boundary

Boundary

Clearly within: binary relevance evaluations with comparably weighted importance of precision and recall. Boundary case: severely imbalanced classes where F1 may hide class-specific performance issues. Clearly outside: evaluations requiring graded relevance, rank-discounted utility (e.g., NDCG), or cost-sensitive weighting where a single harmonic mean is inadequate.

 

 

 

 

 





## Semantic Tension

Semantic Tension

Simplicity (single-number comparability) ↔ Nuance (distinct implications of precision and recall and rank-sensitive utility); F1 trades descriptive nuance for comparability.

 

 

 

 

 





## Synthesis

Synthesis

F1 is a useful compact summary when precision and recall are equally important under binary relevance; it should be interpreted alongside its components and alternative metrics when task objectives, class imbalance, graded relevance, or ranking matters.