Resumen de Support Vector Machines for interval data

Emilio Carrizosa Priego , Frank Plastria , José F. Gordillo Santofimia

• We consider a supervised classi¯cation problem in which the elements to be classi¯ed are balls induced by a symmetric gauge. In particular, this model can be applied to deal with data a®ected by some kind of perturbation and in the case of interval-valued data. Two classi¯cation rules will be de¯ned (a fuzzy and a crisp rule) in terms of a separating hyperplane and a formulation of the problem as a margin maximization model will be introduced, applying this way the standard techniques in Support Vector Machines used for single feature vectors to the case of dealing with balls. We will study in depth the interval data case, where several numerical experiments will be performed. This methodology is also proved to be useful in practice when handling missing values in a database and we replace the missing coordinates by intervals built with the non-missing values of the dataset.