Fourier-Bessel series neural networks for classification

An innovative neural network for classification where hidden neurons are built from double Fourier-Bessel series (FBSNN) is presented. Topology of neural network includes 2 inputs, single output and arbitrary number of hidden neurons. This FBSNN was tested for linearly separable and non-separable, 2- and 3-classes classification problems in machine learning techniques and applications. For comparison, the results obtained from FBSNN were also analyzed by the famous backward propagation neural networks (BP ANN) and radial basis function neural networks (RBFNN), simultaneously. The performance in training process and the decision boundaries of all testing cases are also compared and examined. We found that FBSNN can provides correct classification using small number of hidden neurons and offers very rapid convergence in training process as compared with BPANN and RBFNN.