Maximum Margin Minimum Variance Twin Support Vector Machine

Traditional Support Vector Machine (SVM) based algorithms seek to distinguish two classes by maximizing the margin between two classes without taking into consideration their class distribution information. However, in many practical cases, the distribution of classes plays a crucial role which traditional SVM based classifiers completely ignores. In this paper, we propose a Twin Support Vector Machine based learning model which combines the advantages of generative and discriminative classifiers to learn a robust discriminative model that efficiently considers class-distribution information as well. The resulting classifier has been termed as Maximum Margin Minimum Variance Twin Support Vector Machine $(M^{3}-$ TWSVM). Experimental comparisons of our proposed approach on well-known machine learning benchmark datasets along with real world activity recognition dataset have been carried out. Computational results show that our method is not only fast and yields comparable generalization performance as well.