Maximizing the Area Under ROC Curve with Grouped LASSO

• Activity Number: 342
• Type: Contributed
• Date/Time: Tuesday, August 2, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Sungwoo Choi*+ and Junyong Park
• Companies: University of Maryland at Baltimore County and University of Maryland at Baltimore County
• Keywords: ROC curve ; AUC ; Mann Whitney statistic ; grouped lasso

Abstract

In this talk, we present some results on the estimation of AUC (Area Under ROC curve). In multivariate case, linear combination is a typical way to combine multiple variables and it is incorporated with logistic regression or Mann-Whitney statistic as an objective function. We discuss the optimal score function to maximize AUC and estimate it in high dimension based on selecting meaningful variables by grouped lasso technique. Logistic regression with grouped lasso has been popular whereas Mann-Whitney statistic with grouped lasso has not been widely recognized. We demonstrate the performance of Mann-Whitney statistic with grouped lasso and compare it with logistic with lasso and grouped lasso based on simulations and real data examples.