A novel approach for determining the optimal number of hidden layer neurons for FNN's and its application in data mining

Optimizing the number of hidden layer neurons for an FNN (feedforward neural network) to solve a practical problem remains one of the unsolved tasks in this research area. In this paper we review several mechanisms in the neural networks literature which have been used for determining an optimal number of hidden layer neuron (given an application), propose our new approach based on some mathematical evidence, and apply it in financial data mining. Compared with the existing methods, our new approach is proven (with mathematical justification), and can be easily handled by users from all application fields.