CLASSIFIERS BASED ON A NEW APPROACH TO ESTIMATE THE FISHER SUBSPACE AND THEIR APPLICATIONS

In this thesis we propose a novel classifier, and its extensions, based on a novel estimation of the Fisher Subspace. The proposed classifiers have been developed to deal with high dimensional and highly unbalanced datasets whose cardinality is low. The efficacy of the proposed techniques has been proved by the results achieved on real and synthetic datasets, and by the comparison with state of the art predictors.