Orthogonal linear discriminant analysis and feature selection for micro-array data classification

Abstract

A novel method for micro-array data classification based on orthogonal linear discriminant analysis (ODA), sequential forward floating selection (SFFS) and support vector machine (SVM) is here proposed. In this paper, in order to avoid the constraint that the dimension of the ODA subspace is bounded by the number of classes, to increase the dimension of the subspace and to improve the accuracy, we combine the “original” features to obtain new features. We combine the features in groups of K, each new feature f is obtained by the projection that maps the K-dimensional feature space to a single dimension. A feature selection algorithm is applied to select the most relevant features. Since the new features space has only few hundreds of features an exhaustive wrapper feature selection approach is used to select the set of relevant features. Finally a radial basis function SVM is trained using these features. The obtained results are very encouraging, they improve the average predictive accuracy obtained using standard feature transform techniques. Particularly interesting are the results on a breast cancer dataset, to the best of our knowledge the proposed method is the first method that, using the genes information, permits to determine with high accuracy if a person might benefit from adjuvant chemotherapy.