Does fitcsvm use a kernel by default?

Hello,

I was wondering: when we call fitcsvm to generate an SVM model without specifying a KernelFunction argument, does it find the hyperplane with or without a kernel?

I tried running it without specifying anything and with specifying a linear kernel (which I suspect is either a default MATLAB choice or is the SVM without a kernel) and obtained the same results in term of confusion matrices.

If I understand correctly, there should be a difference between a simple hyperplane cutting through the data in parameter space and the hyperplane/surface cutting through the projection of the data onto a feature space. As the linear kernel is defined as xi'xj (in the manpage for fitcsvm) then this would introduce a mapping/projection, wouldn't it?