An enhanced clustering method for multiple shape basis function networks

By mapping a classifier type fuzzy system into a RBF neural network, tuning of the fuzzy system can be achieved. Using self evolving type RBF networks fuzzy classifiers including the rule base and membership functions can be created. Therefore, it is essential to achieve efficient classification rate in such neural networks. But it is also important to keep the number of automatically added neurons in the hidden layer to a minimum, since those neurons represent the fuzzy rules. This paper introduces a new algorithm of clustering for automatic creation of Multiple Shape Basis Function networks. They can be considered as a generalised form of RBFN with base clusters of different shapes and sizes. The benchmark results show significant improvements and the correct balance between good classification results and the size of the created rule base.