Performance evaluation of hybrid implementation of support vector machine

Abstract

This article focuses on the problem how to shorten the time required for training and decision making by classifiers based on Support Vector Machines techniques. We propose the hybrid implementation of mentioned above algorithm which uses parallel implementation of SVM based on GPU programming model in a distributed computing system using MPI protocol. To estimate the computational efficiency of the proposed model a number of experiments were carried out on the basis of UCI benchmark datasets. Their results show that using parallel model in distributed computing environment can reduce computation time compared to both classical SVM used single processor only and to SVM implementation based on GPU. © 2012 Springer-Verlag.