Feature selection for multi-class classification using pairwise class discriminatory measure and covering concept

An algorithm is presented for selecting a suboptimal set of features which classify given data into classes as effectively as the entire set of features. The algorithm is useful for reducing the number of features in a multi-class problem. In this algorithm, for each pair of classes a feature is successively selected which best discriminates the pair. The algorithm stops when all the pairs are covered. Preliminary experimental results are good