Learning gradual rules to model convex polygon-shaped classes

The work in this paper deals with the learning of gradual rules in the framework of data classification. Gradual rules are well suited to express constraints between numerical quantities. They are here used to constrain the shape of classes to be modeled. More precisely, it is proposed to represent convex polygon-shaped classes by means of "If-Then" classification gradual rules. The latter, learnt from training data, constitute elementary classifiers able to solve oneclass problem with two attributes. General classification problems are thus addressed by combining partial decisions of elementary classifiers. The approach is illustrated with the classification of radar images.