Optimization of GRU Method with Bayesian Optimization for Prediction of South African Rand Exchange Rate against US Dollar

This research compares the performance of the standard Gated Recurrent Unit (GRU) model with GRU optimized using Bayesian Optimization to predict the exchange rate of the South African Rand (ZAR) against the United States Dollar (USD). By utilizing time series data from Yahoo Finance for the period 2018-2023, this research implements a deep learning architecture to capture patterns of currency exchange rate fluctuations. The results show that the GRU model with Bayesian optimization produces better performance on the test data with a MAPE value of 0.81% and R² 0.9352, compared to the standard GRU model with a MAPE of 0.86% and R² 0.9267. Despite the slight decrease in accuracy on the training data, the optimized model has a simpler architecture with a single GRU layer, which indicates better computational efficiency. These findings make a significant contribution to the development of more accurate and efficient currency exchange rate prediction models, particularly for emerging financial markets.