Support Vector Machine (SVM) is a model architecture, originally developed for binary classification, which finds a maximum-margin hyperplane separating the training samples from the two classes. The samples (or vectors) closest to that plane, from either class, are called 'support vectors'.