A novel approach for designing adaptive fuzzy classifiers based on the combination of an artificial immune network and a memetic algorithm

In this study, we propose a novel approach for designing fuzzy classifiers. The first part of our approach is a new preprocess algorithm called SPP (silhouette cluster validity index aided pre-process via k-means). The SPP algorithm has been performed on the data set to determine the numbers of the membership functions and their initial boundaries. Then, the Mopt-aiNetLS algorithm (modified version of opt-aiNet combined with local search strategy of memetic algorithm), the second part of the approach; examines search space to find the optimal values of fuzzy rules and membership functions for the system. The Mopt-aiNetLS is the combination of the memetic algorithm and a modified version of the opt-aiNet algorithm, in which some changes were made in the suppression and hypermutation mechanisms of the original opt-aiNet algorithm. These two new mechanisms are called the intelligent suppression mechanism and the adaptive hypermutation operator. Combining the modified version of opt-aiNet with the local search strategy of the memetic algorithm improves the accuracy of the classification rate. An effective search process has been realized using the Mopt-aiNetLS because the global search capability of opt-aiNet is complemented by the local search strategy of the memetic algorithm. To test the performance of this new approach, twenty different well-known classification dataset benchmark problems from the UCI dataset were used. The average 3 x 10 cross-fold validation results obtained from these datasets are presented and compared with the results of certain classification algorithms reported in the literature. The Wilcoxon Signed-Rank Test was also used for statistical comparisons. The obtained results demonstrate the effectiveness of the proposed approach. (C) 2013 Elsevier Inc. All rights reserved.