SOCP approach to robust twin parametric margin support vector machine

摘要

Twin parametric-margin support vector machine (TPMSVM) is one of the classification tools based on parametric margin ν-SVM (par-ν-SVM) and twin SVM (TWSVM) which is useful for the data having heteroscedastic error structure. However, TPMSVM does not take into account the data uncertainty that might affect model accuracy. In this paper, we study TPMSVM under ellipsoidal uncertainty and its robust counterpart is presented as a second-order cone program (SOCP). Moreover, the robust formulation is extended to the nonlinear case using Gaussian kernels. To evaluate the efficiency of the proposed robust models, experiments have been conducted on nineteen University of California Irvine (UCI) datasets and the results have been compared with TWSVM, twin minimax probability machine classification (TMPMC), TPMSVM, Robust parametric TWSVM (RPTWSVM), and Twin SOCP−SVM (TSOCP−SVM) in the literature. The results show that the new model yieldsbetter classification accuracy.