Artificial Neural Network Technique for Forecasting: Neural Tree and Necessary Number of Hidden Units

This paper introduces a flexible neural tree (FNT) with necessary number of hidden units and is generated initially as a flexible multi-layer feed-forward neural network evolved using an evolutionary procedure and also considers the approximation of sufficiently smooth multivariable function with a multilayer perceptron. For a given neural tree with approximation order, explicit formulas for the necessary number of hidden units and its distributions to the hidden layer of the neural tree are derived. These formulas depend on the number of input variables and on the desired approximation order. The approximation order are widely used in static and dynamic models of nonlinear system. The performance is evaluated using the stock prices. Index terms flexible neural tree, hidden unit, approximation, perceptron, neurons.