F-beta Score
Also known as: F-measure with parameter beta
The F-beta score is a weighted harmonic mean of precision and recall that allows customizing the relative importance of recall versus precision through a parameter beta. It generalizes the F1-score, which is the special case where beta = 1.
Key highlights
- Flexible: allows customization of the precision-recall trade-off based on domain requirements.
- Generalizes the F1-score, providing a family of metrics for different scenarios.
- Clear interpretation: beta directly controls the relative weight given to recall.
- Appropriate for imbalanced datasets with asymmetric error costs.
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Use F-beta score when precision and recall have different importance in your application. Use F2-score (beta = 2) in medical diagnosis where missing a disease is costly. Use F0.5-score (beta = 0.5) in spam filtering where false positives are costly. Choose beta to reflect the relative cost of type II errors (false negatives, penalized by recall) versus type I errors (false positives, penalized by precision).
Strengths & limitations
- Flexible: allows customization of the precision-recall trade-off based on domain requirements.
- Generalizes the F1-score, providing a family of metrics for different scenarios.
- Clear interpretation: beta directly controls the relative weight given to recall.
- Appropriate for imbalanced datasets with asymmetric error costs.
- Requires choosing an appropriate beta value; the choice can be subjective.
- Still ignores true negatives, so does not provide a complete picture of performance.
- Can mask poor performance on specific classes if not combined with per-class analysis.
- More complex to explain to stakeholders than F1-score.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
How do I choose the beta parameter?
Choose beta to reflect the relative cost of false negatives to false positives. If false negatives are twice as costly as false positives, use beta = 2. If false positives are twice as costly as false negatives, use beta = 0.5. If costs are equal, use F1-score (beta = 1). Some practitioners use F2 and F0.5 as standard variants for 'recall-focused' and 'precision-focused' evaluation respectively.
Why not just set different class weights in the learning algorithm instead of using F-beta?
Setting class weights during training optimizes the learned model for specific error costs, while F-beta measures performance on a given model. F-beta is a post-hoc evaluation metric that allows you to assess model performance under different cost assumptions without retraining. However, ideally both should align: use class weights during training and evaluate with matching F-beta.
Is there a way to choose beta automatically from data?
In principle, you could estimate the cost ratio of false negatives to false positives from domain data, but this is often subjective and requires domain expertise. Some practitioners use cross-validation with different beta values to see which provides the best real-world performance, but this requires ground truth data with known costs.
Sources
- 1.van Rijsbergen, C. J. (1979). Information Retrieval (2nd ed.). Butterworth-Heinemann.
- 2.Powers, D. M. (2011). Evaluation: From Precision, Recall and F-Measure to ROC, Informedness, Markedness and Correlation. Journal of Machine Learning Technologies, 2(1), 37-63.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 3). F-beta Score. ScholarGate. https://scholargate.app/model-evaluation/f-beta-score