Independent Component Discriminant Analysis

We introduce a nonparametric method for discriminant analysis based on the search of independent components in a signal (ICDA). Keypoints of the method are reformulation of the classification problem in terms of transform matrices; use of Independent Component Analysis (ICA) to choose a transform matrix so that transformed components are as independent as possible; nonparametric estimation of the density function for each independent component; application of a Bayes rule for class assignment. Convergence of the method is proved and its performance is illustrated on simulated and real data examples.