Seminar Announcement

Generalized Directional Regression: A New Approach for Sufficient Dimension Reduction

Yuexiao Dong

Department of Statistics

Temple University

Jonhson Center 3^rd Floor, Meeting Room B

4400 University Drive, Fairfax, VA 22030

Time: 11 am – 12 pm

Date: Friday, April 6, 2012 

Abstract

As a popular dimension reduction technique, directional regression achieves good accuracy by implicitly synthesizing the first two conditional moments. In this paper, we extend directional regression to a general family of estimators via the notion of general empirical directions. Our proposed estimators inherit the nice properties of directional regression, such as unbiasedness, exhaustiveness, and square root n consistency. Cross validation is used to choose the best estimator within this family. Simulation results show that our proposed cross validation based general directional regression estimator achieves similar performance as the oracle classification rule with categorical response, and has good overall performance when the response is continuous. An empirical study with the wine recognition data set further demonstrates the improvement of general directional regression over classical directional regression.