Hybrid Method for Fast SVM Training in Applications Involving Large Volumes of Data

Abstract

One of the problems of training a Support Vector Machine (SVM) for applications involving large volumes of data is how to solve the constrained quadratic programming issue. The optimization process suffers from the problem of large memory requirement and computation time. In this paper we propose a hybrid genetic algorithm based SVM that addresses the large memory requirement and computation time problem. The system operates in two main stages. During first stage it obtains a subset of features using genetic algorithm and during second stage it uses genetic algorithm to train the SVM using subset of features. The proposed system is tested on gene expression profile data sets. The experiment results show that the proposed hybrid system is efficient from memory and time computational point of views without compromising classification accuracy results.