Comparison of neural networks and regression time series on forecasting development of US imports from the PRC

Abstract

The contribution deals primarily with the development and prediction of the PRC export to the USA, comparing traditional statistical methods in the form of regression analysis of time series and artificial neural networks, which are a very important prediction tools, and become an integral part of modelling and predicting certain development of time series based on the statistical data. The USA import from China can be defined based on the statistical, causal, and intuitive methods. In this case, the contribution deals primarily with comparing the statistical methods. The contribution provides only a possible framework of the monitored variable development. It is necessary to work also with the information about the future economic, political or legal environment. If it is possible to predict their development, it can then be reflected in the monitored variable. Optically, the best option from linear regression appears to the curve obtained by the least squares method by negative exponential smoothing. As for the neural networks, all retained structures appear to be applicable in practice. In terms of the correlation coefficient, only neural networks are applicable.

Keywords

• regression time series
• neural networks
• prediction
• import