Area under the Precision-Recall Curve (AUC-PR)

This chapter focuses on investigating the area under the precision-recall curve (AUC-PR). In addition to the Summary, there are four sections. Section 18.1 introduces six alternative approaches to using the area under the receiver operating characteristic curve (AUC-ROC): (1) avoid using AUC-ROC, (2) modifying AUC-ROC, (3) use performance metrics that do not rely on observed absences, (4) use adaptive AUC-ROC methods, (5) use cost curves on AUC-ROC, and (6) use AUC-PR. Section 18.2 introduces the AUC-PR by defining PR and ROC curves and describing the relationship between PR and ROC curves. Section 18.3 demonstrates how to calculate the AUC-ROC and AUC-PR values using the three R packages SIAMCAT, PRROC, and PRECREC . Section 18.4 provides remarks on AUC-PR that describe the differences between the AUC-PR and AUC-ROC (Section 18.4.1), how AUC-PR leverages the advantages of ROC and PR curves (Section 18.4.2), the advantages of AUC-PR over AUC-ROC (Section 18.4.3), particularly explaining why AUC- PR outperforms AUC-ROC in machine learning of microbiome statistics (Section 18.4.4), and the disadvantages of AUC-PR (Section 18.4.5).