Weighted intuitionistic fuzzy least squares twin SVM

Abstract

Twin support vector machine is one of the widely used classifiers for real world applications. Many algorithms are proposed in the literature to deal with noisy data sets. One of the different approaches is the intuitionistic fuzzy twin support vector machine (IFTWSVM) technique. However, most of the real-world problems have data sets with some correlation between two points in the same class which is not addressed by IFTWSVM. Within this thesis, we propose a twin SVM based classifier for binary data sets termed as weighted intuitionistic fuzzy least squares twin support vector machines. The proposed model not only judges input data by considering the membership and nonmembership values but also uses the correlation between two data points in the same class for the betterment of classification performance. This weighting technique can give weights considering the relative significance of each data point. Retrospecting the time complexity for big data sets we have used the least square model, which can solve the classification problem without solving any Quadratic Programming Problem. We have checked the efficiency of the proposed model on a considerable amount of data sets. It can be seen that the proposed model can generalize brilliantly when compared to other baseline models in terms of accuracy and area under the curve (AUC).