Predicting Gold Prices Using N-BEATS and DBN: A Deep Learning Perspective
Daniel Makala, Zongmin Li · 2025
This study evaluates the predictive performance of two advanced models—N-BEATS (Neural Basis Expansion Analysis for Time Series) and Deep Belief Networks (DBN)—in forecasting gold prices. Given the importance of gold as a financial asset, accurate price prediction is vital for investors and market analysts. Both models were tested on a dataset of historical gold prices and their performances were assessed using key metrics RMSE, MAE, MAPE, and R². The results indicated that N-BEATS outperformed DBN in three of the four metrics. Specifically, N-BEATS recorded an RMSE of 21.06, an MAE of 16.06, and a MAPE of 0.79%, while achieving an R² value of 0.99. In comparison, DBN achieved an RMSE of 21.61, an MAE of 16.14, a MAPE of 0.80%, and an identical R² of 0.99. Although both models demonstrated high accuracy in terms of R², N-BEATS exhibited superior performance in RMSE, MAE, and MAPE, suggesting a lower average magnitude of error in predictions. These findings highlight the efficacy of N-BEATS as a robust model for forecasting gold prices, offering better predictive accuracy and interpretability than DBN.