Receiver Operator Characteristic Tools Graphic User Interface Extension for R Commander

Abstract

Background: Receiver Operator Characteristic (ROC) curve, is a graphical plot which presents the performance of a binary classifier when the discrimination cutoff is varied. The aim of this work was to create an extension for R Commander that offers a graphical user interface for Receiver Operator Characteristic tools provided by several existing command line accessible packages like pROC and ROCR. Material and Methods: The extension was built and tested with R version 3.2.0 and R Commander 2.1-7. Results: We built an extension called RcmdrPlugin.ROC that we uploaded on the CRAN servers. The extension adds a new menu called ROC, along with two submenus pROC and ROCR that broadly corresponds to commands available to access the functions of these packages. The pROC menu offers several commands: to plot a ROC curve for a dataset or for a logistic regression model, to compare paired and unpaired ROC curves, each providing the following tabs: General (to select the variables for the analysis, and options for switching cases with controls); Smoothing (allowing the user to select different types of smoothing – binominal, density, distributions like normal, lognormal, ...); AUC (to specify the partial area under the curve (AUC) options), CI (to select the options of confidence intervals (CI) – the level, computing method: DeLong, bootstrap, ...); Plot (for the plotting options). The ROCR dialogue window offers more options in choosing the performance measures for the plot. Conclusion: The RcmdrPlugin.ROC extension helps less advanced users of R accessing ROC tools in a friendly graphical user interface.