A New Parameter Selection Method for Support Vector Machine Based on the Decision Value

Abstract

Abstract To overcome the disadvantage of CV-ACC method that the high-density sample region may be close to the optimal hyper-plane, a parameter selection method for support vector machine (SVM) based on the decision value, named as CV-SNRMDV method, is proposed in this paper. SNRMDV is used as the criterion of cross-validation (CV) in our method, which is defined as the ratio between the difference of medians of decision values and the sum of the standard deviations from the medians. Compared with the traditional cross-validation accuracy (CV-ACC) method, CV-SNRMDV makes use of the information of sample distribution and decision value. Consequently CV-SNRMDV overcomes the disadvantage of CV-ACC. The experiments show our method obtains a better test accuracy on the simulated dataset, while the test accuracies on benchmark datasets are close to CV-ACC.