It selects support vectors based on distance metrics.

It reduces the dataset to a single dimension to simplify computations.

 It is used only for linear SVMs to compute the margin.

It allows computation of inner products in a high-dimensional space without explicitly mapping points.