Using Ant Colony Optimization for Learning Maximal Structure Fuzzy Rules

Usually, the rules in a fuzzy model contain in the antecedent a set of propositions each of which restricts a fuzzy variable to a single fuzzy value by means of the predicate equal-to. That way, each rule covers a single fuzzy region of the fuzzy grid. This paper proposes to extent this structure in order to provide more general fuzzy rules, in the sense of covering the input space as much as possible. In order to do this, new predicates are considered and an ant colony optimization algorithm is proposed to learn such fuzzy rules. The obtained fuzzy models provide two benefits: they are described with a lower number of rules and their accuracy improves with the increase in generalization introduced. Some experimental results illustrate these facts