Sensitivity analysis in linear programming approach to optimal SVM classification

Abstract

At present, linear programming (LP) techniques for optimal one-class and two-class classification can be considered well established and feasible; they pose an alternative to the quadratic programming (QP) approach, which is usually credited with having greater complexity. Sensitivity analysis, well developed in the LP context, is generally employed to furnish answers describing how an optimal solution changes when varying the parameters in an LP problem; as a possible application in optimal classification, it can be employed for the definition of the free parameters present in LP procedures, reducing the events of computational restart from scratch when searching for a satisfactory classifier through repeated trials. The proposed method is demonstrated on a simple example which exhibits its effectiveness in reducing the computational burden, but this procedure can be extrapolated to large problems as well. Keywords : Linear Programming, Optimal Classification, Sensitivity Analysis, Support Vector Machines. Normal 0 14 false false false IT X-NONE X-NONE