Feature Selection Based On Linear Twin Support Vector Machines.

Abstract: By promoting the parallel hyperplanes to non-parallel ones in SVM, twin support vector machines (TWSVM) has been attracted wildly attention. However, the SVM feature selection algorithm (such as SVM-RFE) cannot be used to TWSVM directly. In this paper, we propose two TWSVM feature selection algorithms for classification problems. Firstly, by analyzing the weights in classification, we merge the two weights of the non-parallel hyperplanes in linear TWSVM into one, and propose the sort-TWSVM feature selection by sorting the merged weight; Secondly, inspire by SVM-RFE, we propose the TWSVM-RFE feature selection in a similar way with SVM-RFE by using the merged weight. Preliminary experiments on several benchmark datasets show the feasible and effective of our sort-TWSVM and TWSVM-RFE on feature selection. [Copyright &y& Elsevier]