Data points are separable by a linear classifier with a guaranteed minimum distance (margin) from the decision boundary.