Comparing CNN and GRU for Gold Price Prediction Using Deep Learning

This research proposes a hybrid deep learning model that integrates Convolutional Neural Networks (CNNs) and Gated Recurrent Units (GRUs) to predict gold prices. The motivation stems from the volatile and complex nature of the gold market, heavily influenced by macroeconomic indicators such as the exchange rate (IDR/USD), Bank Indonesia (BI) interest rate, and inflation. In the hybrid architecture, the CNN serves as a feature extractor to identify nonlinear patterns in historical and economic data. At the same time, the GRU captures temporal dependencies, enabling the model to learn both short-term and long-term dynamics. The dataset comprises daily gold prices from January 2020 to August 2024, enriched with macroeconomic indicators to improve predictive relevance. Experimental results show rapid convergence of training and validation losses within 12 epochs. Model evaluation using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Mean Absolute Percentage Error (MAPE) demonstrates high predictive accuracy, with a MAPE of 1.136%. A comparative analysis with standalone CNN and GRU models reveals that the hybrid CNN–GRU architecture consistently outperforms both in terms of accuracy and prediction stability. This study contributes to financial forecasting by providing a robust, data-driven predictive tool that can support timely investment decisions in volatile market conditions.