Selection of best bases for classification and regression

Abstract

We describe extensions to the "best-basis" method to select orthonormal bases suitable for signal classification (or regression) problems from a collection of orthonormal bases using the relative entropy (or regression errors). Once these bases are selected, the most significant coordinates are fed into a traditional classifier (or regression method) such as linear discriminant analysis (LDA) or a classification and regression tree (CART). The performance of these statistical methods is enhanced since the proposed methods reduce the dimensionality of the problems by using the basis functions which are well-localized in the time-frequency plane as feature extractors.