A New Hybrid Method for Identification of Fuzzy Models
Pietari Pulkkinen, Hannu J. Koivisto · 2006
The aim is to develop a method capable of identifying the adequate structure and parameters of fuzzy models (FMs) by combining initialization algorithms, simplification methods and genetic algorithm (GA). Fuzzy function estimators and classifiers are initialized by modified Gath-Geva (MGG) and C4.5 algorithms, respectively. Then, a 3-step GA optimization is performed. During it, simplification operators, extended with a new rule's antecedents reducing method, are performed and simple FMs can be rewarded by a new fitness function. Several classification and function estimation problems are studied. Comparisons of the obtained models with models in the literature show promising results in terms of interpretability, compactness and accuracy.