Learning fuzzy rules for modeling complex classification systems using genetic algorithms

Genetic algorithms, as general purpose learning techniques, have been widely applied in the modeling of fuzzy rules-based classification system. However, the algorithms are more vulnerable to local convergence as a result of the increasing complexity and dimensionality of classification problems, which reduces the performance of the algorithms. To prevent the algorithms only learning rules from small subset of the search space, a fitness sharing method based on the similarity level of one rule from its neighbours rules is proposed. The similarity level is calculated by the similarity values of different antecedent fuzzy sets, which are cached for reducing the additional computing load. The proposed method is studied for two complex data, the sonar signals classification and the hand movement recognition problems. And the experimental results demonstrate that the proposed method is able to efficiently achieve accurate performance.