Area Under the Precision-Recall Curve

Measure to compare true observed labels with predicted probabilities in binary classification tasks.

Usage

prauc(truth, prob, positive, na_value = NaN, ...)

Arguments

truth

(factor())
True (observed) labels. Must have the exactly same two levels and the same length as response.

prob

(numeric())
Predicted probability for positive class. Must have exactly same length as truth.

positive

(character(1))
Name of the positive class.

na_value

(numeric(1))
Value that should be returned if the measure is not defined for the input (as described in the note). Default is NaN.

...

(any)
Additional arguments. Currently ignored.

Value

Performance value as numeric(1).

Details

Computes the area under the Precision-Recall curve (PRC). The PRC can be interpreted as the relationship between precision and recall (sensitivity), and is considered to be a more appropriate measure for unbalanced datasets than the ROC curve. The PRC is computed by integration of the piecewise function.

This measure is undefined if the true values are either all positive or all negative.

Meta Information

• Type: "binary"

• Range: \([0, 1]\)

• Minimize: FALSE

• Required prediction: prob

References

Davis J, Goadrich M (2006). “The relationship between precision-recall and ROC curves.” In Proceedings of the 23rd International Conference on Machine Learning. ISBN 9781595933836.

See also

Other Binary Classification Measures: auc(), bbrier(), dor(), fbeta(), fdr(), fnr(), fn(), fomr(), fpr(), fp(), mcc(), npv(), ppv(), tnr(), tn(), tpr(), tp()

Examples

truth = factor(c("a", "a", "a", "b"))
prob = c(.6, .7, .1, .4)
prauc(truth, prob, "a")
#> [1] 0.904106