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

This study performs a comparative examination of three deep learning models: Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and Artificial Neural Network (ANN) to predict the USD/NGN exchange rate utilizing a decade-long historical dataset obtained from Investing.com. The dataset underwent thorough pre-processing and normalization, subsequently followed by an 80/20 train-validation division. The performance of each model was evaluated using conventional regression metrics: Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and R-squared (R²). Of the three models, the GRU exhibited the most performance, attaining the lowest MAE (0.609), the lowest RMSE (0.103), and a commendable R² score of 0.629. These findings indicate that GRU provides a superior approach for short-term exchange rate forecasting, especially in turbulent financial markets such as that of the Nigerian Naira.