Learning and optimization of fuzzy rule base by means of self-adaptive genetic algorithm

This work presents an approach for automatic fuzzy rule base generation and optimization by means of self-adaptive genetic algorithm, that changes dynamically the crossover and mutation rates ensuring population diversity and avoiding premature convergence. The application domain is multidimensional fuzzy pattern classification, where the class also is fuzzy. The membership functions were defined by the fuzzy clustering algorithm FC-Means. We first describe the fuzzy rules format and fuzzy reasoning method for pattern classification problems. After this, the genetic fuzzy rule base learning from given examples based on Pittsburgh approach implemented here is introduced. Next the genetic fuzzy rule base optimization process used to exclude unnecessary and redundant rules is described. The performance of our method is evaluated on some well-known data sets. Compact fuzzy rule bases were generated with high classification ability. The dynamic change of crossover and mutation parameters showed that great improvement can be achieved to results. The use of "don't care" condition allows to generate more comprehensible and compact rules.