Feature Decorrelation for Variable Selection Using a Constrained Whitening Approach (306559)Francesca Chiaromonte, The Pennsylvania State University and Sant’Anna School of Advanced Studies
*Ana Maria Kenney, The Pennsylvania State University
Keywords: Feature Selection, Regression, Whitening
Strong correlations among features are well-known hurdles for existing variable selection/screening methods. Previous studies demonstrated that transforming predictors through a pre-processing step called ZCA whitening can greatly improve accuracy in certain selection procedures. However, this whitening method induces complete de-correlation at the cost of similarity with the original set of predictors. We propose a novel approach that only induces de-correlation up to a point while maintaining a pre-specified level of similarity, or distance, between the new and original variables. We demonstrate the benefits and drawbacks of this method when applied prior to various selection techniques through an in-depth simulation study and real data application.