AUCprecisionrecall

Area under the precision-recall curve (AUPRC) is a summary statistic used to evaluate binary classifiers across all decision thresholds. It measures the trade-off between precision (the proportion of true positives among predicted positives) and recall (the proportion of true positives identified among all actual positives) as the threshold varies. Compared with the area under the ROC curve (AUROC), AUPRC places more emphasis on the positive class and is especially informative when the positive class is rare.

Calculation typically starts with a model that outputs scores or probabilities. Precision and recall are computed at various thresholds by ranking instances from highest to lowest score and moving the threshold accordingly. The resulting precision-recall curve is then integrated to yield the AUPRC. The integral is commonly approximated with numerical methods such as the trapezoidal rule. In some contexts, the related measure average precision (AP) is used, and for monotonic curves AP often equals the AUPRC.

Interpretation of AUPRC values: they range from 0 to 1, with 1 indicating perfect discrimination. The baseline is the positive class prevalence; a random classifier yields an AUPRC approximately equal to this prevalence. AUPRC can be sensitive to class distribution, and values are not directly comparable across datasets with different rates of positive instances.

Applications and considerations: AUPRC is widely used in binary classification evaluation, particularly in imbalanced settings such as medical risk scoring, fraud detection, and information retrieval. For multiclass problems, AUPRC can be computed per class in a one-vs-rest approach or averaged using macro or micro methods.