A study on generalization capability of weighted fuzzy production rules with maximum entropy

For enhancing the representation power of fuzzy production rules (FPRs), weighted fuzzy production rules (WFPRs) are considered by incorporating the concept of weight into FPRs. This paper investigates the weights´ impact on the generalization capability of WFPRs. Given a fact and a set of WFPRs, a reasoning conclusion which can be drawn by matching the fact against the set of WFPRs is dependent of the weight values of WFPRs. Adjusting the weight values can lead to a change of the reasoning conclusion, and therefore, can lead to a change of generalization capability of WFPRs. For a given dataset from which a set of FPRs are extracted, this paper proposes to determine the weight values based on the well known maximum entropy principle (MEP). Initial experiments show that the inclusion of weights determined according to MEP can result in an improvement of generalization capability of WFPRs for selected databases.