search for




 

TROC curve and accuracy measures
Journal of the Korean Data & Information Science Society 2018;29:861-72
Published online July 31, 2018
© 2018 Korean Data and Information Science Society.

Chong Sun Hong1 · Su Jin Lee2

12Department of Statistics, Sungkyunkwan University
Correspondence to: Professor, Department of Statistics, Sungkyunkwan University, 25-2, Sungkyunkwan-Ro, Jongno-Gu, Seoul 03063, Korea. E-mail: cshong@skku.edu
Received May 29, 2018; Revised June 28, 2018; Accepted July 2, 2018.
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/3.0) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
Abstract
ROC curve is a useful graphical method to visualize the classifier based on performance and evaluate its accuracy. Since ROC curve does not reflect sample sizes of the two groups, there is a problem in evaluating the classifier’s performance when the samples are imbalanced severely. TOC curve was proposed to overcome these shortcomings. But it is difficult to compare and describe the performance of the classifier since TOC curve is implemented as a parallelogram. In this study, we propose TROC curve that overcomes the disadvantages of TOC curve. Also TROC curve could preserve the usefulness of ROC curve and express the degree of imbalance of the sample. By using TROC curve, it is possible to grasp the information on the classification result more easily. Various accuracy measures are described based on TROC curve similar to ROC curve, and can be expressed better than TOC curve. Through simulation and illustrative examples, we discuss that TROC curve can visualize and identify imbalanced data, and can explain more information than ROC and TOC curves easily.
Keywords : Accuracy, classifier, confusion, default, threshold.