F1score

F1 score, or F1 measure, is a metric used to evaluate classification performance, especially in binary and multilabel settings. It is the harmonic mean of precision and recall. Precision P = TP/(TP+FP) and recall R = TP/(TP+FN). The F1 score is defined as F1 = 2PR/(P+R) = 2TP/(2TP+FP+FN). It can be computed from a confusion matrix.

In multiclass or multilabel contexts, F1 can be extended through micro-averaging, macro-averaging, or weighted averaging. Micro-F1 aggregates TP, FP, and FN across all classes before applying the formula, while macro-F1 computes a per-class F1 and averages the results, and weighted-F1 averages per-class F1 weighted by the class’s support. F1 is particularly useful when there is class imbalance and a balance between precision and recall is desired, rather than overall accuracy.

Limitations and interpretation: F1 does not consider true negatives, which can be misleading in imbalanced datasets or when the negative class dominates. It also depends on the threshold used to convert predicted scores into labels. The F-beta family generalizes F1 by weighting recall more (β > 1) or precision more (β < 1), with F1 corresponding to β = 1.