New encoding scheme for evolving fuzzy classifiers

We present a noble encoding method for designing an optimal fuzzy classifier with evolutionary optimization. Evolutionary designs of fuzzy classifiers is divided into design of fuzzy rules and design of fuzzy membership functions. Among these design problems, for an evolutionary design of membership functions, the shapes of each membership function are mainly considered in the previous related works. In other words, design of fuzzy membership functions is formulated as a parameter search problem for tuning the shapes of each function (e.g. center(or mean) and width(or variance) in a Gaussian function). In this paper, we newly consider the design of fuzzy membership functions as optimization of intersection positions between adjacent membership functions. According to recent insightful researches, classification boundaries are determined by the points of intersection of membership functions. Therefore, the proposed approach differs from conventional approaches in that the proposed method can search and manipulate the border of classification which directly influences the classification performance. In order to verify the proposed encoding method, simulation study is carried out. For this simulation study, we apply the proposed encoding scheme to the basic genetic algorithm (GA), one of the most widely used evolutionary optimization methods in the recent literatures. The performance of the proposed method is investigated with two real world databases, dasiairispsila and dasiaglasspsila data.