Comparative Analysis of LSTM, GRU, and ANN Models for USD/NGN Exchange Rate Forecasting

Abstract

Deep learning models have the capability to analyse extensive historical and real-time Forex data, identifying intricate patterns and relationships that can help predict future market movements. Furthermore, these models can assess sentiment derived from news sources and social media to gauge market dynamics, while also automating trading decisions to enhance performance. By harnessing deep learning, traders strive to achieve a competitive advantage through improved predictive accuracy and more efficient risk management. The objective of this research is to construct a model for accurately forecasting the exchange rate of the United States Dollar (USD) against the Nigerian Naira (NGN) using three machine learning algorithms: Long Short-Term Memory (LSTM), Gated Recurrent Units (GRU), and Artificial Neural Network (ANN). These algorithms were selected due to their demonstrated effectiveness in prior studies. This research explores machine learning fundamentals, its applications, various methodologies, and algorithms. Additionally, the study outlines different feature extraction techniques employed in the analysis. For this research, data was sourced from the Investing.com dataset and divided into training and testing subsets. The model was trained and evaluated using the specified machine learning techniques. To assess its predictive accuracy and efficiency, metrics such as Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE) were applied. The findings revealed that GRU outperformed both LSTM and ANN, with an MAE of 0.609, an RMSE of 0.103, and an