| Popis: |
Background. Generation of a consistent base of fuzzy production rules is an important task, but no less important is the task of determining the most efficient generation algorithm. The purpose of this work is to research the effectiveness of six algorithms for generating a base of fuzzy production rules for the Wang-Mendel neural network using the example of solving a classification problem. Materials and methods. All researches werecarried out in the environment of the developed application, in which the software model of the Wang-Mendel network is implemented. The following algorithms have been studied for generating the rule base: an adaptive learning algorithm, the Abe-Len algorithm, a rating algorithm for generating fuzzy production rules, C-Means and C-Ellipses clustering algorithms, and new hybrid algorithm proposed in this paper and based on the Abe-Len algorithm and the rating algorithm. The researches have been done on the basis of four model data sets from the UCI repository: Fisher irises, types of wine drinks, types of breast cancer, and «Space Shuttle» data. Results. The research showed that the hybrid Abe-Len algorithm in combination with the rating algorithm and by the C-Ellipse clustering algorithm provides the best results. At the same time, the C-Ellipse algorithm assumes a manual preliminary selection of the optimal number of clusters, while this is not required in the proposed modification of the Abe-Len algorithm. Conclusions. The conducted research and implemented algorithms made it possible to obtain numerical estimates of the error in solving the classification problem and justify the choice of the most effective algorithms for generating a fuzzy rule base. The combination of the Abe-Len algorithms and the rating algorithm provides the maximum efficiency on the taken data sets. |
