IMCP: A Python package for imbalanced and multiclass data classifier performance comparison

Abstract

The Multiclass Classification Performance (MCP) curve is an innovative method to visualize the performance of a classifier for multiclass datasets. On the other hand, the Imbalanced Multiclass Classification Performance (IMCP) curve is a novel approach to visualizing classifier performance on multiclass datasets that exhibit class imbalance, i.e. the proportions of (two or more) class labels are unequal. We have developed an open-source Python package that encompasses the functionality required to calculate and visualize these two novel classification performance measures, along with providing the calculation of the area under the curves. The MCP and IMCP curves offer advantages over the traditional ROC (Receiver Operating Characteristic) curve when dealing with multiclass and imbalanced datasets, respectively. They provide more informative insights into classifier behavior, especially in scenarios involving multiple classes or uneven class distribution.