Between AUC based and error rate based learning
Kar‐Ann Toh · 2008
Based on an earlier solution to optimize an approximated area under the ROC Curve (AUC) for binary pattern classification in [1], this paper investigates into the relationship between AUC and several error rate based classifiers. Via a generalized framework of translated scalingspace, we find that the AUC based classifier can be related to a total-error-rate (TER) classifier, an Equal Error Rate (EER) formulation, and a least-squares-error (LSE) estimator, each under a specific setting of the translated scaling-space framework. Several potential applications of the generalized framework are subsequently discussed.