arXiv:2404.13147v1 Announce Kind: new
Summary: Mannequin analysis is of essential significance in trendy statistics utility. The development of ROC and calculation of AUC have been broadly used for binary classification analysis. Current analysis generalizing the ROC/AUC evaluation to multi-class classification has issues in at the least one of many 4 areas: 1. failure to supply wise plots 2. being delicate to imbalanced knowledge 3. unable to specify mis-classification price and 4. unable to supply analysis uncertainty quantification. Borrowing from a binomial matrix factorization mannequin, we offer an analysis metric summarizing the pair-wise multi-class True Constructive Price (TPR) and False Constructive Price (FPR) with one-dimensional vector illustration. Visualization on the illustration vector measures the relative velocity of increment between TPR and FPR throughout all of the lessons pairs, which in turns offers a ROC plot for the multi-class counterpart. An integration over these factorized vector offers a binary AUC-equivalent abstract on the classifier efficiency. Mis-clasification weights specification and bootstrapped confidence interval are additionally enabled to accommodate a wide range of of analysis standards. To help our findings, we carried out in depth simulation research and in contrast our technique to the pair-wise averaged AUC statistics on benchmark datasets.
Supply hyperlink