A deterministic approach to regularized linear discriminant analysis

Abstract

The regularized linear discriminant analysis (RLDA) technique is one of the popular methods for dimensionality reduction used for small sample size problems. In this technique, regularization parameter is conventionally computed using a cross-validation procedure. In this paper, we propose a deterministic way of computing the regularization parameter in RLDA for small sample size problem. The computational cost of the proposed deterministic RLDA is significantly less than the cross-validation based RLDA technique. The deterministic RLDA technique is also compared with other popular techniques on a number of datasets and favorable results are obtained.