Classification problems with imprecise data through separating hyperplanes
Emilio Carrizosa, José Manuel Gordillo, Frank Plastria · 2007
We consider a supervised classification problem in which the elements to be classified are sets with certain geometrical properties. In particular, this model can be applied to deal with data aected by some kind of noise and in the case of interval-valued data. Two classification rules, a fuzzy one and a crisp one, are defined in terms of a separating hyperplane, and a formulation of the rule identification problem by margin maximization is introduced, extending the standard techniques in Support Vector Machines used for single feature vectors. We study in depth the interval data case and report on several numerical experiments. This methodology is also proved to be useful in practice when handling missing values in a database.