Local Hybrid SVMDT Classifier

Abstract

Support vector machines are among the most precise classifiers available, but this precision comes at the cost of speed. There have been many ideas and implementations for improving the speed of support vector machines. While most of the existing methods focus on reducing the number of support vectors in order to gain speed, our approach additionally focuses on reducing the number of samples, which need to be classified by the support vector machines in order to reach the final decision about a sample class. In this paper we propose a novel architecture that integrates decision trees and local SVM classifiers for binary classification. Results show that there is a significant improvement in speed with little or no compromise to classification accuracy.