The Community for Technology Leaders
Green Image
We present an algorithm which provides the one-dimensional subspace where the Bayeserror is minimized for the C class problem with homoscedastic Gaussian distributions. Ourmain result shows that the set of possible one-dimensional spaces v, for which the order ofthe projected class means is identical, defines a convex region with associated convex Bayeserror function g(v). This allows for the minimization of the error function using standardconvex optimization algorithms. Our algorithm is then extended to the minimization of theBayes error in the more general case of heteroscedastic distributions. This is done by meansof an appropriate kernel mapping function. This result is further extended to obtain the d-dimensional solution for any given d, by iteratively applying our algorithm to the null space ofthe (d — 1)-dimensional solution. We also show how this result can be used to improve uponthe outcomes provided by existing algorithms, and derive a low-computational cost, linearapproximation. Extensive experimental validations are provided to demonstrate the use ofthese algorithms in classification, data analysis and visualization.
Linear discriminant analysis, feature extraction, Bayes optimal, convex optimization, pattern recognition, data mining, data visualization

A. M. Martinez and O. C. Hamsici, "Bayes Optimality in Linear Discriminant Analysis," in IEEE Transactions on Pattern Analysis & Machine Intelligence, vol. 30, no. , pp. 647-657, 2007.
84 ms
(Ver 3.3 (11022016))