Um Novo Algoritmo de Treinamento para Máquinas de Vetores Suporte

Abstract

We propose a training algorithm for support vector machines. Our proposal is based the on decision function of support vector machines, which is a kind of distance measure. We use such a kind of distance measure to select the support vectors as well as set their Lagrange multiplier values. We compare our proposal with sequential minimal optimization (SMO) and classical quadratic optimization problem (QP) solver in terms of accuracy, precision, recall and training time for several datasets. In general, the results are equivalent in accuracy; however, our proposal is faster than SMO to reach model during the training process for all the available datasets. Besides that, we highlight that our proposal can handle datasets with large number of patterns in much less time when compared to sequential minimal optimization and classical quadratic optimization problem (QP) solvers.