B1225
Title: Variable selection in AUC-optimizing classification
Authors: Seung Jun Shin - Korea University (Korea, South) [presenting]
Abstract: Optimizing the receiver operating characteristics (ROC) curve is often desired in imbalanced classification. A binary classifier is proposed that optimizes the area under the ROC curve (AUC) penalty and is penalized by the smoothly clipped absolute deviance (SCAD) penalty, referred to as the SCAD-AUC estimator, and its properties are thoroughly studied. The SCAD-AUC estimator is established to possess the oracle property in high dimension, enabling the proposal of a consistent BIC-type information criterion that greatly facilitates the tuning procedure. Both simulated and real data analyses demonstrate the promising performance of the proposed SCAD-penalized AUC-optimizing classifier in terms of variable selection and prediction.
