Performance evaluation of various variants of fuzzy classifier systems for pattern classification problems
Hisao Ishibuchi, Tomoharu Nakashima · 2002
We have already proposed a fuzzy classifier system for efficiently generating fuzzy if-then rules from numerical data for high dimensional pattern classification problems with many continuous attributes (H. Ishibuchi et al., 1995; 1996). We examine the performance of various variants of our fuzzy classifier systems by computer simulations on commonly used real world test problems. Those variants are implemented in the following manners: (i) using a different coding method; (ii) combining a learning procedure of each fuzzy if-then rule with the fuzzy classifier system; (iii) introducing a heuristic procedure for generating an initial population of fuzzy if-then rules; (iv) introducing a heuristic procedure for generating new fuzzy if-then rules that are used for replacing the worst rules in the current population; (v) extending the constant population size to an adjustable parameter; (vi) introducing the concept of voting for combining multiple fuzzy rule bases.