Universum Twin Bounded Support Vector Machine Via Augmented Lagrangian Algorithm and Its Application

In this paper, we propose an augmented Lagrangian algorithm for the twin bounded support vector machine with universum data (UTBSVM), which is designed for solving the optimization problem in primal space. By using the augmented Lagrangian method, we convert the constrained problems into unconstrained problems. So, the subproblems in the augmented Lagrangian method reduces to non-smooth subroplems, that by using smoothing technique that convert into smooth subproblems. Moreover, the smooth reformulations of subproblems, called AL-SUTBSVM, are solved by a modification of Newton’s algorithm. The experiment result on a handwritten digit recognition data set demonstrate that our method is more effective than some popular methods in aspect of classification accuracy.

کلیدواژه ها:

Universum ، Twin bounded support vector machine ، Augmented Lagrangian method ، Smoothing technique ، Newton’s method.