Non-linear Regression Based on the Analytic Center of Feasible Region

Non-linear regression algorithm is proposed based on the analytic center of feasible region, which overcomes the drawback the generalization performance of support vector regression degrades when the feasible space is elongated or asymmetric; the proof is given that the algorithm proposed is equivalent to the maximum likelihood parameter estimation when fitting error satisfies a class of probability density function. The experiments demonstrate the validity of the non-linear regression proposed in this paper