Support vector machines (SVMs) belong to the family of the kernel methods.
The main features of SVM are the following:
• It is a supervised learning method which can be used for classification, regression, and outliers’ detection.
• It exploits the “kernel trick” for distances in order to extend linear techniques to the solution of non-linear problems.
• It is effective in high dimensional spaces and uses a subset of training points in the decision function (called support vectors), so it is also memory efficient.
• As kernel methods, it selects the model which minimizes the trade-off between the performance over the data and the complexity of the solution.
• The main hyperparameters of the SVM algorithm are the kernel’s type (which is usually the Gaussian one for kernel SVM), kernel’s hyper-parameter γ, and the regularization’s parameter usually indicated as C or α. The regularization parameter controls the complexity of the solution, while the kernel’s hyperparameter controls the non-linearity of the solution.
• SVM does not handle categorical features directly (consequently encoding is needed).
• SVM suffers from numerical issues and consequently input data must be scaled.

Christianini, N. and Shawe-Taylor, J. C., An Introduction to Support Vector Machines and Other Kernel-Based Learning Methods, Cambridge University Press, 2000.

Giulia Cademartori

University of Genoa