F2Werte

F2-Werte, often referred to as the F2-Score in German-language texts, are a metric used to evaluate binary classification models. They belong to the family of F-measures and place greater emphasis on recall than on precision, making them suitable when missing positive cases is more costly than generating false positives.

The Fβ score is defined from precision and recall. Let TP be true positives, FP false positives, and FN false negatives. Precision P = TP / (TP + FP) and recall R = TP / (TP + FN). The Fβ score is Fβ = (1 + β²) · P · R / (β² · P + R). For β = 2, the F2 value is F2 = 5PR / (4P + R). The F2 value ranges from 0 to 1, with higher values indicating better performance under the weighting that favors recall.

Calculation typically uses the confusion matrix from a classifier’s predictions. After computing P and R from TP, FP, and FN, the F2 value provides a single-number summary that reflects the balance between precision and recall, biased toward recall. It is common to adjust decision thresholds to optimize F2 for a given application, especially in domains where identifying all positive cases is critical.

Limitations and context: F2-Werte should not be interpreted in isolation or compared across unrelated tasks with different class distributions. They depend on the chosen threshold and the relative costs of FP and FN. Other Fβ measures (for example F1, F0.5) reflect different priorities, so the choice of β should align with domain-specific goals.