Minimum Class Variance SVM+ for data classification

In this paper, a new Support Vector Machine Plus (SVM+) type model called Minimum Class Variance SVM+ (MCVSVM+) is presented. Similar to SVM+, the proposed model utilizes the group information in the training data. We show that MCVSVM+ has both the advantages of SVM+ and Minimum Class Variance Support Vector Machine (MCVSVM). That is, MCVSVM+ not only considers class distribution characteristics in its optimization problem but also utilizes the additional information (i.e. group information) hidden in the data, in contrast to SVM+ that takes into consideration only the samples that are in the class boundaries. The experimental results demonstrate the validity and advantage of the new model compared with the standard SVM, SVM+ and MCVSVM.