Accessing Deep Learning Techniques for Commodity Price Prediction: A Comparative Assessment Across Energy, Precious Metal, and Agricultural Futures

This study performs an in-depth comparison of deep learning approaches - CNN, LSTM, GRU, and their bidirectional variants (BiLSTM and BiGRU) - to anticipate commodity futures prices, particularly in the crude oil, gold, and corn markets. The approach used in our study combines fundamental and technical factors with thorough variable selection to establish the most effective predictors. The Grid Search Cross-Validation approach is employed for optimizing the hyperparameters of deep learning architectures. The efficacy of applied models is evaluated using common evaluation indicators: Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Coefficient of Determination(R2), and Adjusted R2. The results presented enhance the existing research on AI applications in financial markets, providing recommendations for the development of more precise forecasting techniques for researchers.