An analytic center machine for regression

Support vector machines represent a new approach for solving problems in pattern classification and regression analysis. Because of their impressive generalization peformance they have attracted much attention in the optimization and machine learning communities. In the version space of hypotheses the optimal support vector machine solution corresponds to the computation of the Chebyshev center. Recent research efforts, however, have suggested that the selection of alternative centers further improves the generalization performance. We propose an algorithm for regression analysis that finds the hypothesis that corresponds to the analytic center of the version space. Preliminary results indicate that a higher level of generalization accuracy can be achieved by this regression estimator.