Support vector regression with local ϵ parameters with the support vectors

In support vector machine regression (SVR) a big ε value gives a rough system model with little support vectors and a small ε value gives an accurate system model with many support vectors. The selection of the support vectors is effected by a small change of the training data. To obtain an accurate model with little support vectors, a method includes two steps is proposed in this paper, in step one, a big ε value is used to select a small number of the support vectors; in step two, by giving these selected support vectors a small value while others a big one, a accurate system model is obtained. The experimental results demonstrate the efficiency of the proposed method.