An Introduction to Bidirectional Recurrent Neural Networks in Forecasting Commodity Prices

An Introduction to Bidirectional Recurrent Neural Networks in Forecasting Commodity Prices

연구 분야: Artificial Intelligence

학회: International Conference on Asia Pacific Advanced Network

This paper investigates the predictive capabilities of Bidirectional Recurrent Neural Networks (BRNNs) in forecasting commodity prices, specifically focusing on Brent oil. Unlike traditional models such as ARIMA and GARCH, which primarily capture linear relationships and often struggle with market volatility, BRNNs process sequences in both forward and backward directions. This bidirectional approach enables BRNNs to capture complex, non-linear dependencies within the data, making them better suited for the dynamic and unpredictable nature of commodity markets. The performance of the BRNN model was rigorously evaluated against ARIMA and GARCH models using key metrics. The BRNN model achieved a Mean Absolute Error (MAE) with a significant lower than ARIMA’s and GARCH’s. Similarly, the Root Mean Square Error (RMSE) and The Mean Absolute Percentage Error (MAPE) for BRNN stood at 25.05%, substantially outperforming ARIMA’s 32.45% and GARCH’s 34.12%. These results demonstrate the BRNN model’s superior predictive accuracy and its enhanced ability to respond to market volatility. This study not only establishes the BRNN model’s dominance over conventional methods but also offers a novel approach that significantly improves predictive accuracy in volatile markets. The findings have practical implications for traders, analysts, and policymakers, providing a more reliable and advanced tool for commodity price forecasting.