The support vector machine is a linear discriminant classifer. SVM separates the class entities by a linear discriminant line. The fatter the margin between the discriminant and the class entity and the better the lower the penalty for error for class entity on the wrong side then the better the prediction.