Long-Term Gold Price Prediction Using the CNN-BiGRU Model Enhanced with an Attention Mechanism

Gold has been regarded as a precious metal by humankind since ancient times, and its unique physical and chemical properties have made it important in several fields, including industry, finance, and jewelry manufacturing. As a safe-haven asset, fluctuations in the price of gold directly impact the risk and return of investment portfolios, underscoring the significance of accurate forecasting for investors, policymakers, and financial institutions. Conventional time series forecasting methodologies are inadequate in capturing the intricate nonlinear patterns exhibited by gold price fluctuations. In this paper, we propose a hybrid deep learning model, CNN-BiGRU-AM, with the aim of enhancing the accuracy of gold price predictions. This hybrid model integrates the strengths of both CNN and BiGRU, leveraging CNN to identify significant features in historical price data and BiGRU to handle long-term dependencies in time-series data. Additionally, it incorporates an attention mechanism, a key innovation that enhances the model's predictive accuracy. The test results show that, on the same dataset, the proposed model makes the most accurate predictions compared to the other four models. It has an RMSE value of 0.9811, which is much lower than the CNN-GRU-AM (RMSE value of 1.085) and GRU (RMSE value of 1.8623).