Reordenación de los ejemplos en fuzzy-ARTMAP para ganar robustez y disminuir el número de categorías: Aplicación a la extracción automática de reglas.

Abstract

One of the drawbacks for the fuzzy-ARTMAP algorithm is its dependence on the order of the training samples. When we want to extract linguistic rules from samples, we need a system able to produce as few rules as posible, while maintaining its liability. We propose that an overcome of the difficulty can arise from a recursive reordering of the training samples acording with parameters suggested by the algorithm itself.