Simultaneous classification and feature selection via LOG SVM and Elastic LOG SVM

In data mining and machine learning, classifying the class labels and selecting features simultaneously are important. This study proposes two new sparse support vector machines (SVMs), namely, LOG SVM and Elastic LOG SVM. The LOG SVM uses the LOG penalty, and the Elastic LOG SVM combines the non-convex LOG penalty and the L2 norm penalty. The LOG SVM and Elastic LOG SVM can achieve classification and feature selection simultaneously. Local quadratic approximation is used to solve both SVMs. Experiments are also conducted to show that the proposed SVMs perform well in the aspects of classification and feature selection.