Robust Variable Selection Based on the Density Power Divergence Loss

• Activity Number: 505
• Type: Contributed
• Date/Time: Wednesday, August 3, 2016 : 8:30 AM to 10:20 AM
• Sponsor: International Chinese Statistical Association
• Author(s): Yang Li* and Wenfu Xu and Yichen Qin and Shuangge Ma
• Companies: and Renmin University of China and University of Cincinnati and Yale University
• Keywords: robust variable selection ; density power divergence ; adaptive lasso

Abstract

The minimum density power divergence (DPD) criterion is an useful method for robust estimation. In order to deal with the data with contamination, we propose a robust variable selection method with the DPD loss and adaptive lasso penalty. Numerical studies shows that the proposed method, compared with the traditional regularization method with OLS or MLE, is resistant to outliers on both the performance of variable selection and estimation. Oracle properties and root n consistence are proved with reasonable regularity conditions. Real data examples are analyzed to illustrate the performance of the proposed method. And it shows that the proposed method is not congruent with other recent procedures, emphasizing the importance of applying and developing robust variable selection methods.