Abstract

Fishers linear discriminant analysis (LDA) is a classical multivariate technique both for dimension reduction and classication. The data vectors are transformed into a low dimensional subspace such that the class centroids are spread out as much as possible. In this subspace LDA works as a simple prototype classier. The resulting decision boundaries are linear. However, in many applications the linear boundaries do not adequately separate the classes and the possibility of modeling more complex boundaries would be desirable. In this paper we present a nonlinear generalization of discriminant analysis that implements the method of representing dot products of pattern vectors by kernel functions. This technique allows to eciently compute discriminant functions in arbitrary feature spaces for which such kernel representations exist. 1 Introduction Classical linear discriminant analysis (LDA) is a statistical method that attempts to project data vectors that belong to c dierent classes...

Citations