Closest Pairs Data Selection for Support Vector Machines

Chaofan Sun

Abstract: This paper presents data selection procedures for support vector machines (SVM). The purpose of data selection is to reduce the dataset by eliminating as many non support vectors (non-SVs) as possible. Based on the fact that support vectors (SVs) are those vectors close to the decision boundary, data selection keeps only the closest pair vectors of opposite classes. The selected dataset will replace the full dataset as the training component for any standard SVM algorithm.