A framework for optimising fuzzy inference in classifier systems

In generating a suitable fuzzy classifier system, significant effort is placed on the determination and the fine tuning of the fuzzy sets. In such systems, little thought is given to the selection of the most suitable inference strategy. Often a traditional inference strategy is applied which allows no control over how strong or weak the inference is applied. A number of theoretical fuzzy inference operators have been proposed but not investigated in real world applications. This paper proposes a novel genetic algorithm framework for optimizing the strengths and weaknesses of fuzzy inference operators concurrently with a set of membership functions for a given fuzzy classifier system. The paper investigates several theoretical proven fuzzy inference techniques and applies them within the proposed framework. The results from three real world data sets establish that the choice of inference parameters has a significant effect on the accuracy and robustness of fuzzy classifiers.