Review of SVM Applications in Chemistry

The fact that the linear kernel gives better results than nonlinear kernels for certain QSAR problems is documented in the literature. Yang et al. compared linear, polynomial, and RBF kernels for the following properties of alkyl benzenes boiling point, enthalpy of vaporization at the boiling point, critical temperature, critical pressure, and critical volume. A LOO test showed that the first four properties were predicted best with a linear kernel, whereas critical volume was predicted best with a polynomial kernel. 

---