Extensión de la SVM para regresión a tramos

Abstract

In this paper, a new technique for unidimensional regression based on support vector machines is presented. The method is based on the addition of new linear restrictions to the standard (or global) support vector machines technique. Despite the new linear restrictions, a quadratic programming problem is also obtained. Moreover, the use of support vector machines allows us a straightforward derivation of a nonlinear extension of the proposed technique. Finally, simulation results are presented in which the proposed technique is applied to an equalization problem, in this example the proposed technique presents similar results to those of the Bayesian (optimal) equalizer. The technique is also applied to a nonlinear regression problem in which local modeling provides better results than global modeling.