A More Computationally Efficient Model Selection Method for Regularized Discriminant Analysis

John A. Ramey, Phil D. Young, and Dean M. Young.
Working Paper, Baylor University.

Availability:

Abstract. Regularized discriminant analysis (RDA) is a widely used supervised classification rule that performs well in the high dimensional, small sample size (p >> N) classification setting. However, one of its disadvantages is that the proposed model selection method is computationally intensive and, therefore, is often impractical. In this paper we propose a heuristic method that reduces the computational burden and has good classificatory performance. We use the expected error rate (EER) to assess the performance of our proposed model selection method and compare it to the grid model selection method of Friedman (1989). We find that our heuristic method has excellent results for a variety of simulation configurations and consistently reduces the computational burden required in RDA.

Keywords: Regularized Discriminant Analysis, High-Dimensional Classification, Shrinkage Estimation, Model Selection.

BibTeX Record:

@TechReport{ramey10efficientrda,
  author       = {John A. Ramey and Phil D. Young and Dean M. Young},
  title        = {A More Computationally Efficient Model Selection Method for Regularized Discriminant Analysis}
  year         = 2010,
  institution  = {Baylor University},
  type         = {ERID Working Paper},
  number       = 50
}