COMBINING NEURAL NETWORKS DURING TRAINING FOR REAL TIME SERIES MODELING AND FORECASTING

Neural networks nn's have been widely used as nonlinear models for time series. recently, many researchers have performed a combination of several nn's in order to attain a better accuracy model. in this paper a number of combination models are developed during the training of the neural networks .the best performer amongst those networks when tested on a validation data set is selected as the eventual combination model. one step ahead forecasting of real life data sets using the combined model is performed. comparisons are made between the two techniques in combining an ensemble of neural networks .the new proposed technique in combining the neural networks during training cdt resulted in smaller forecasting errors for the real life time series under consideration, hence showing a better fit and understanding to the nature of the data.