Neyman-Pearson Criterion (NPC): a Model Selection Criterion for Asymmetric Binary Classification

• Activity Number: 229 - Advances in the Neyman-Pearson Classification
• Type: Topic Contributed
• Date/Time: Monday, July 29, 2019 : 2:00 PM to 3:50 PM
• Sponsor: WNAR
• Author(s): Yiling Chen* and Jingyi Jessica Li and Xin Tong
• Companies: University of California, Los Angeles and University of California, Los Angeles and University of Southern California
• Keywords: model selection; disease prediction; type I error; false negative rate; asymmetric errors; Neyman-Pearson criterion

Abstract

Cancer diagnosis based on genomic features belongs to the application of binary classification where the two types of classification errors do not have the same priority. Given that, the overall classification error should not be a proper measure of prediction accuracy, and therefore the genomic features selected based on this criterion might fail to serve medical research and practices. Motivated by the Neyman-Pearson (NP) binary classification paradigm that prioritizes the error of higher importance, we propose a new model selection criterion: Neyman-Pearson Criterion (NPC). Theoretical model selection property of NPC is studied for nonparametric plug-in methods. Simulation studies show that NPC outperforms the criterion of minimizing the overall classification error when practitioners have a desirable upper bound on the type I error. A real data application to a breast cancer DNA methylation dataset suggests that NPC is a practical criterion that can reveal novel gene markers for cancer diagnosis with both high sensitivity and specificity.