Designing RBF neural networks with weighted mean subtractive clustering algorithms

In this paper, weighted mean subtractive clustering algorithms are proposed to find cluster centers of the dataset. Then the found cluster centers act as the centers of radial basis functions. In weighted mean subtractive clustering algorithms, subtractive clustering is used to find center prototypes and then weighted mean methods are used to create new centers. Three weighted mean methods are tried to create more effective centers. Comparative experiments were executed between subtractive clustering and three weighted mean subtractive clustering algorithms on five benchmark datasets. Next, the performance of RBF neural networks set with the proposed algorithms was studied. The experimental results suggest that all three weighted mean subtractive clustering algorithms can find more accurate centers and can be successfully applied to design RBF neural networks. The RBF neural networks determined by weighted mean subtractive clustering algorithms have rather simpler network architecture but with slightly lower classification accuracy than ones determined by subtractive clustering algorithm.