SVM is a supervised classification method based in a kernel geometrical construction, as shown in the following diagram. SVM can be applied for either classification or regression, because a classification problem can be treated as a special type of regression problem, assuming that each observation is placed into one, and only one, of the categories of the values of the predictors. SVM will look for the best decision boundary that splits the points into the classes they belong to. To accomplish this SVM, we will look for the largest margin (space free of training samples parallel to the decision boundary).

In the following diagram, we can see the margin as the space between the dividing line and dotted lines, which extend support vector classifiers to accommodate nonlinear class boundaries. SVM will always look for a global solution because the algorithm only cares about the vectors close to the decision boundary. The points in the edge of the margin are the support...