Using the leader algorithm with support vector machines for large data sets

Abstract

One of the main drawbacks of Support Vector Machines (SVM) is their high computational cost for large data sets.We propose the use of the Leader algorithm as a preprocessing procedure for SVM with large data sets, so that the obtained leaders are used as the training set for the SVM. The result is an algorithm where the Leader algorithm allows to construct a sample of the data set whose granularity level and computational cost are controlled by the threshold parameter. Despite its apparent simplicity, the proposed model obtains similar accuracies to standard LIBSVM with fewer number of support vectors and less execution times. © 2011 Springer-Verlag.