Time series forecasting with neural network ensembles: an application for exchange rate prediction

摘要: This paper investigates the use of neural network combining methods to improve time series forecasting performance traditional single keep-the-best (KTB) model. The ensemble are applied difficult problem exchange rate forecasting. Two general approaches networks proposed and examined in predicting between British pound US dollar. Specifically, we propose systematic serial partitioning build ensembles for It is found that basic approach created with non-varying architectures trained using different initial random weights not effective improving accuracy prediction while models consisting structures can consistently outperform predictions ‘best’ network. Results also show based on partitions data more than those developed full training out-of-sample Moreover, reducing correlation among forecasts made by members utilizing techniques key success models. Although our considerable advantages over KTB approach, they do have significant improvement compared widely used walk model