PRROC: computing and visualizing precision-recall and receiver operating characteristic curves in R.

Area of Science:

• Machine Learning
• Bioinformatics
• Computational Statistics

Background:

• Classifier performance evaluation is crucial in machine learning and bioinformatics.
• Precision-Recall (PR) and Receiver Operating Characteristic (ROC) curves are standard metrics.
• Existing R packages have limitations in handling soft-labeled data for these curves.

Purpose of the Study:

• To introduce the R-package PRROC for computing and visualizing PR and ROC curves.
• To provide a method for calculating PR and ROC curves with continuous interpolation for soft-labeled data.
• To offer a generic plotting function for publication-quality PR and ROC curve graphics.

Main Methods:

• Development of the R-package PRROC.
• Implementation of continuous interpolation for PR curve points.
• Integration of area under the curve (AUC) calculations for both PR and ROC curves.
• Inclusion of a generic plotting function for enhanced visualization.

Main Results:

• PRROC enables the computation and visualization of PR and ROC curves.
• The package accurately handles soft-labeled data through continuous interpolation.
• AUC for PR and ROC curves can be computed efficiently.
• Publication-quality graphics for PR and ROC curves are generated.

Conclusions:

• The PRROC R-package provides a valuable tool for classifier performance assessment.
• It addresses limitations of existing packages by supporting soft-labeled data.
• PRROC facilitates the generation of high-quality visualizations for research publications.