Structure optimization of fuzzy neural network using rough set theory
Jung-Heum Yen, Seung-Moo Yang, Hong-Tae Jeon · 1999
This paper presents an approach to obtain a reduced neuro-fuzzy model for a plant. The reduction is carried out through an iterative algorithm aiming to select a minimal number of rules of the model. To decide which rules we may eliminate, dependency in rough set theory is used. Dependency between each rule in a model and the output of the plant allows one to see how much contribution the rule has to the identification of the plant. While the reduced model maintains the same performance as the original one, the selection algorithm can minimize its complexity and redundancy of the structure. The rapid convergence of the number of the redundant rules must be accomplished by our method. One does not need to cluster the input space from the raw data and, furthermore, ignores the /spl epsiv/-completeness which has to be considered when adjusting the membership functions. Experimental results demonstrate the effectiveness of using dependency to measure the contribution of any rule to the model.