Neuro-fuzzy network with dynamic adjustment in forecasting

Abstract A neuro-fuzzy network is employed to model forecasting evenly. In order to obtain high precision and high adaptability in forecasting, the modeling method consists of the following phases: 1) using a BP network to select input variables from candidate inputs; 2) building a neuro-fuzzy network by means of fuzzy clustering; 3) conducting parameter identification by enhanced learning BP algorithm; 4) dynamically adjusting the structure and parameters of the neuro fuzzy network. To verify the performance of the network, this neuro-fuzzy network is used to predict the sales of a company. The conventional ARIMA model and neural model are used for the comparison purpose. The result shows that the proposed forecasting method using neuro-fuzzy network achieves higher precision and adaptability