First of all, you should use the regression framework for SVM, i.e., Support Vector Regression (SVR), as far as you are trying to predict a continous variable (the frequency), not a classification. And, obviously, you need some training data consisting on different antenna parameters and the resulting resonant frequency.
This kind of model will work only if you're using always the same type of antenna (e.g., a patch), and you have properly characterized the parameters that have an effect on the resonant frequency (e.g., size of each dimension, thickness, dielectric constant of the substrate...; they depend on the type of antenna you're considering).
Take into account that you will need these training data in advanced, so you will probably need to get them through full-wave simulation, electromagnetic analysis, etc., typically using a software tool. Once you have enough data, just let the SVR get trained... and let the magic flow ;-)
Just one note: be aware that it will only work provided that you have properly choiced the antenna parameters, and that the relation bewteen such parameters and resonant frequency exists and is unique. A Machine Learning tool cannot solve ambiguity, so uniqueness is a must.