ADVANCED MULTI-TASK TWIN SUPPORT VECTOR MACHINE WITH UNIVERSUM DATA FOR MEDICAL ANALYSIS

Multi-task learning (MTL) is an emerging field in machine learning that aims to enhance the performance of multiple interconnected learning tasks by leveraging the shared information among them. Drawing inspiration from MTL, a recent advancement known as the multi-task twin support vector machine with Universum data (UMTSVM) has been introduced. In this paper, we present a novel approach to tackle the UMTSVM problem, referred to as NUMTSVM. Our approach involves transforming the related optimization problems into unconstrained optimization problems. The objective functions of these unconstrained problems are piecewise quadratic and only once differentiable. Consequently, we propose a modified version of Newton’s method to effectively solve these unconstrained problems. Experimental evaluations conducted on medical data sets demonstrate the superior efficiency and accuracy of our proposed NUMTSVM method compared to other existing methods in the literature.