An efficient and user-friendly software tool for ordered multi-class receiver operating characteristic analysis based on python

Abstract

Receiver Operating Characteristic (ROC) analysis is a prevalent tool for two-class problems in a wide range of science and engineering areas. The relevant scalar area under the ROC curve (AUC) has been proven to be a robust evaluation index for characterizing the performance of binary statistical models. However, tasks with more than two categories are frequently encountered in practice. As an extension of the ROC curve and AUC in parallel, the hyper ROC surface and the corresponding hyper-volume under the surface (HVUS) have attracted extensive interest. The current mainstream analysis tools of AUC and HVUS are based on the R program language. To provide researchers with an alternative tool for multi-class ROC analysis, in this work, we develop a python-based package and software tool, which allows the users to unbiasedly estimate HVUS, confidently compare HVUSs, and flexibly visualize 2D-ROC curve and 3D-ROC surface, etc.

Keywords

• Receiver Operating Characteristics (ROC)
• Hyper-volume under the surface (HVUS)
• Efficient
• Python