Extensions to Regularised Discriminant Analysis

Regularised Discriminant Analysis has proven to be a most eeective classiier for problems where traditional classiiers fail because of a lack of suucient training samples, as is often the case in high dimensional settings. However, it has been shown that the model selection procedure of Regularised Discriminant Analysis, determining the degree of regu-larisation, has some deeciencies associated with it. We propose a modiied model selection procedure based on a new appreciation function. By means of an extensive simulation it was shown that the new model selection procedure performs better than the original one. We also propose that one of the control parameters of Regularised Discriminant Analysis be allowed to take on negative values. This extension leads to an improved performance in certain situations. The results are connrmed using two chemical data sets.