Threshold single multiplicative neuron model artificial neural networks for forecasting problem

Single multiplicative neuron models artificial neural networks, which are frequently used in time series forecasting problems and do not have any restriction on the number of hidden layer unit, have an important place in artificial neural networks literature. In the studies about single multiplicative neuron models artificial neural networks, the model obtained from training process is a single model. In this thesis, differently from many other artificial neural network models in the literature, a threshold value is determined to obtain the output of the system of single multiplicative neuron models artificial neural networks. According to obtained threshold value, it is determined which weights and biases are used for the calculation of the output of the network. It has been separately used harmony search algorithm and particle swarm optimization for training the single multiplicative neuron model based on a threshold value, which is used to determine the optimal weights and bias of the system. To evaluate the performance of the proposed method, three different real life time series were analysed and the results obtained from both the methods' proposed in this thesis and many popular artificial neural network methods in the literature were compared.