Support Vector Machines (page 4 of 10)

The resulting trained SVM model may have a hard or a soft margin. Hard Margin refers to that kind of decision boundary that makes sure that all the observations are classified correctly. While hard margins minimize error, they often cause the margins to shrink, which minimizes the usefulness of the SVM algorithm. Soft Margin refers to having observations that are mis-specified due to regularization. Soft margins allow us to better balance bias and variance.