A comparison of feed forward and recurrent neural networks in time series forecasting

Abstract Exchange rate prediction has been treated as a key research matter in international finance because it’s a crucial and an important subject. For this reason, many researchers have sought to develop an accurate technique to forecast series of exchange rates. In this same context, this paper targets at making a comparison between two neural networks architectures, such as RNN( Recurrent Neural Networks) and FFNN (Feed Forward Neural Networks), to forecast the series of exchange rates as USD/NOK, USD/EUR for a daily frequency extending from november 2017 throught november 2022 either 1305 observations. In order to evaluate and compare the quality and performance of these approaches, we have resorted to MSE( Mean Square Error). The back propagation method is used as the basis for developing algorithms capable of forecasting the series of exchange rates. To our knowledge, this is the only paper that focuses on the comparison between two techniques of neural networks, such as RNN and FFNN for predicting exchange rates. The results clearly show that the recurrent neural network approach (RNN) yielded the highest prediction accuracy, and it’s more robust and more efficient in forecasting compared to the Feed Forward Neural Network model (FFNN). It’s, therefore, the most recommended.