Fast Training of Support Vector Machines by Extracting Boundary Data

Abstract

Support vector machines have gotten wide acceptance for their high generalization ability for real world applications. But the major drawback is slow training for classification problems with a large number of training data. To overcome this problem, in this paper, we discuss extracting boundary data from the training data and train the support vector machine using only these data. Namely, for each training datum we calculate the Mahalanobis distances and extract those data that are misclassified by the Mahalanobis distances or that have small relative differences of the Mahalanobis distances. We demonstrate the effectiveness of the method for the benchmark data sets.