Support Vector Regression is a well established robust method for function estimation. The Support Vector Machine uses inner-product kernels between support vectors and the input vectors to transform the nonlinear classification and regressions problem to a linear version.function where the surface is approximated with a linear
combination of the kernel function evaluated at the support vectors. In many applications, the number of these support vectors can be quite large which can increase the length of the prediction phase for large data sets. Here we study a technique for reducing the number of support vectors to achieve comparable function estimation accuracy. The method identifies support vectors that are close to the ε-tube and uses them to approximate the function estimate of the original algorithm.