Optimizing Gated Recurrent Unit (GRU) for Gold Price Prediction: Hyperparameter Tuning and Model Evaluation on Historical XAU/USD Data

This study investigates the use of a Gated Recurrent Unit (GRU) model with a four-layer architecture for daily gold price closing prediction, motivated by the model's ability to effectively capture temporal dependencies in time series data. Gold price forecasting is highly challenging due to its volatility and external factors, making it an important area of research for investors and financial analysts. By systematically optimizing hyperparameters through 72 combinations of epochs, batch size, GRU layer units, and dropout rates, the study identifies the optimal configuration (100 epochs, batch size of 16, 256 units, dropout rate 0.1) based on MSE performance on validation data. The best model achieved MAE of 25.76, MSE of 954.97, and RMSE of 30.90, after inverse transformation on test data. These results highlight the potential of the GRU model in accurately forecasting gold prices, with implications for financial decision-making . However, the prediction error suggests that further improvements could be made by incorporating external factors or exploring advanced model architectures.