Abstract

CARDONA M., CESAR AUGUSTO  and  VELASQUEZ H., JUAN DAVID. SELECTION OF EXCELLENT CHARACTERISTICS USING MUTUAL INFORMATION. Dyna rev.fac.nac.minas [online]. 2006, vol.73, n.149, pp.149-163. ISSN 0012-7353.

A new method for relevant attribute selection based on Mutual Information is presented. It is based on the concept of relevance probability of each attribute, which is measured using a permutation test, and it allows to discard irrelevant variables thus as to order by importance the relevant variables. The proposed methodology is tested using three well know classification problems. Also, a research is conducted to clarify its performance when the relevant variables are contained with noise, or there are irrelevant artificial random variables. The results show the success of the proposed methodology, by which it is recommended its use as integral part of the tools used in relevant attribute selection

Keywords : Mutual Information; Classifiers; Attribute Selection.

        · abstract in Spanish     · text in Spanish     · Spanish ( pdf )