search for


Unresolved inference using ROC-based reject rule
Journal of the Korean Data & Information Science Society 2018;29:1319-28
Published online September 30, 2018
© 2018 Korean Data and Information Science Society.

Chong Sun Hong1 · Sol Mi Park2

12Department of Statistics, Sungkyunkwan University
Correspondence to: Professor, Department of Statistics, Sungkyunkwan University, Seoul 03063, Korea. E-mail:
Received August 30, 2018; Revised September 17, 2018; Accepted September 17, 2018.
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License ( which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
In the process of evaluating the performance of the classi fication model for two distribution functions based on the cost function, unreliable samples were rejected based on the ROC-based reject rule due to the high cost of wrong classifi cation in order to minimize classi fication errors in the classi fication process. In this paper, we consider rejected samples between the two rejection thresholds obtained by the ROC-based reject rule as the unresolved, so we propose a method for estimating the optimal threshold by setting a new cost function. In de ning the probability density functions for the unresolved, various weights are given only to the cost of the misclassfi ed FN and FP in order to reset the cost function for the unresolved. The changes in cost corresponding to the optimal threshold minimizing the cost function are then examined. For the samples between the two rejection thresholds, which are considered to be the unresolved, the new cost functions are divided into the cases where the costs of FN and FP are the same and the cases where the costs are different. In the cost of the unresolved, the results differ according to changes in the misclassifi cation cost. However, the optimal threshold of the unresolved is similar to that of the estimated optimal threshold with respect to various weights of the misclassi fications.
Keywords : Cost, misclassi fication, reject rule, threshold, unresolved.