Double-layer Support Vector Machine for Robust and Efficient Classification

Abstract

In this paper, double-layer support vector machine (DLSVM) for binary classification is proposed that can expeditiously learn a classification model under the premise of ensuring accuracy. In the first layer, least squares support vector machine (LSSVM) is used to evaluate all samples in the data set. Support vector machine (SVM) utilize a sparse data set to train the binary classifier in the second layer. In order to obtain excellent sparse data set for SVM, the normalized norm sequence and the normalized Lagrange multiplier Shannon entropy sequence are introduced. DLSVM provides low computation cost because the two-layer structure reduces lots of iterative operations. Experimental results show that the proposed method not only has similar classification accuracy compared to SVM but also has higher efficiency.