Assigning local weights within fuzzy production rules for improving reasoning accuracy

When a set of fuzzy production rules which are acquired by learning from training examples have poor reasoning accuracy with respect to the training examples, one may use a refining method to improve the reasoning accuracy. The paper proposes a new approach to refine the fuzzy production rules, which assigns local weights to propositions of fuzzy production rules by using a linear programming procedure. In addition to the reasoning accuracy improvement, this approach has a number of advantages such as intuitive background of local weights, non-increasing of number of rules, and less computational effort for obtaining local weights