A Study On Parametric Sensitivity Of Fuzzy Reasoning Based On Similarity

Fuzzy production rule reasoning based on similarity is a kind of method generally used in classification. It depends on the selection of related functions, such as membership, similarity and match operator. Different functions maybe lead to different reasoning results. Thus, many experts study on the mathematical properties of these functions and their application to reasoning. But most of the work only focuses on the comparison of different types of functions in reasoning, few on the sensitivity of parameters in the same type function.This paper analyzes the sensitivity of parameters in some functions, including Gaussian-type membership function, Minkowski distance and Yager's t-norm and t-conorm, which are used in the similarity-based reasoning frequently. On the basis of theoretical and experimental analysis, it gives some guidelines on how to select appropriate values of the parameters. This study on sensitivity is useful to improving the accuracy, reasonability and generality of reasoning. It also has practical values in improving the intelligence of inference system.