Nonlinear discriminant analysis based on probability estimation by gaussian mixture model

Abstract

The Bayesian a posterior probability is a very important element in pattern recognition. In classification problems, the posterior probabilities reflect the uncertainty of assessing an example to particular class. Such residual information will be useful for more deep understanding or analysis of examples. In this paper, we propose a nonlinear discriminant analysis based on the probabilistic estimation of the Gaussian mixture model (GMM). We use GMM to estimate the Bayesian a posterior probabilities of any classification problems. Then we use posterior probabilities estimated by GMM to construct discriminative kernel function. The performance of the proposed kernel function is confirmed by several experiments using UCI machine learning repository. © 2014 Springer-Verlag Berlin Heidelberg.