A parallel classification and feature reduction method for biomedical applications

Abstract

Classification is one of the most widely used methods in data mining, with numerous applications in biomedicine. The scope and the resolution of data involved in many real life applications require very efficient implementations of classification methods, developed to run on parallel or distributed computational systems. In this study we describe SVD-ReGEC, a fully parallel implementation, for distributed memory multicomputers, of a classification algorithm with a feature reduction. The classification is based on Regularized Generalized Eigenvalue Classifier (ReGEC) and the preprocessing stage is a filter method algorithm based on Singular Value Decomposition (SVD), that reduces the dimension of the space in which classification is accomplished. The implementation is tested on random datasets and results are discussed using standard parameters. © 2008 Springer-Verlag Berlin Heidelberg.