Evaluation of Performance of MLP Neural Networks and RBF Neural Networks in Adjusting Time Series of the Development of the Trade Balance Between the USA and the PRC

Abstract: Artificial neural networks are widely used for predicting values, for solving possible future problems and are able to provide various solutions in problem estimation, regression or optimisation. They are useful for predicting time series too. The aim of the paper is to analyse and evaluate the performance of multilayer neural networks (hereinafter referred to as “MLP”) and neural networks of radial basis function (hereinafter referred to as “RBF) in adjusting time series on the example of the trade balance between the United States and the People’s Republic of China. Regression was performed using neural structures. We generated multilayer perceptron networks and neural networks of radial basis function and we generated two sets of artificial neural networks. Time was the continuous independent variable. We determined the trade balance of the USA and the PRC as a dependent variable. We can state that due to the great simplification of reality, it is not possible to predict the emergence of extraordinary situations and their impact on the trade balance of the USA and the PRC. We can state that when an adjusted time series is derived from a single variable, time, RBFs perform better than MLPs. In order to make the prediction more accurate and its calculation easier, it seems appropriate to use RBF networks, which brings a relatively high degree of accuracy.