Check the libsvm page for how to train SVM to learn non$-$linear decision boundaries such as a Polynomial or an RBF curve.

Build a non$-$linear SVM model with a polynomial kernel with degree $2$ & test it on the test set.

Repeat with a polynomial kernel with degree $3$

Repeat with an RBF kernel with gamma $=0\mathrm{.}01$

Repeat with an RBF kernel with gamma $=0\mathrm{.}1$