An empirical assessment on the robustness of support vector regression with different kernels
Jingxu Liu, Jin Li, Yuejin Tan · 2005
The choice of kernels is important for the support vector regression (SVR). In this paper, the robustness of SVR with different kernels is empirically analyzed. The two typical kernels, polynomial kernel and radial basis function (RBF) kernel, and their hybrid are used. Two simple rules for composition of kernels are used to produce the hybrid kernels. The experimental results show that the SVR with different scale kernels of the same type has different performance. The SVR using the polynomial kernel with lower degree are more robust than that using RBF kernel with narrower parameter. The hybrid of different types or scales of kernels can improve the robustness to some extent. Furthermore, two benchmark datasets are used to verify the results.