3.2 Support Vector Machines
Support vector machines (SVM) are rooted in statistical learning theory [2, 12].
We consider SVM for the classification task formalized in section 2. The space
H of all possible functions for SVM is the set generated by all possible oriented
hyperplanes in an n-dimensional Euclidian space Rn or in a higher dimensional
feature space F obtained by a mapping φ(x) on the instances x from Rn. SVM
search for an oriented hyperplane that results in a maximal margin between the
two classes, while minimizing the penalty term for the training instances at the
wrong side of the margin.
Given l training instances xi, SVM solve the following primal problem: