<![CDATA[This study compares the performance of Long Short-Term Memory (LSTM) and Temporal Fusion Transformer (TFT) in forecasting daily gold prices using multivariate data. The dataset was obtained from Kaggle (2005–2024) and includes ten key economic variables, such as stock indices, the US Dollar Index, crude oil prices, silver prices, and 10-year Treasury yields. The research stages consisted of data preprocessing through missing value interpolation, Z-score-based outlier clipping, normalization with MinMaxScaler on the training set, and data transformation tailored to each model architecture. Model performance was evaluated using four regression metrics: Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), R², and Mean Absolute Percentage Error (MAPE). Results indicate that TFT outperforms LSTM across all metrics, achieving RMSE of 19.35, MAE of 14.51, R² of 0.9906, and MAPE of 0.74%. The Diebold–Mariano (DM) test yielded a p-value of 0.02, confirming that the performance difference between the two models is statistically significant. These findings highlight the importance of the attention mechanism and variable selection network in TFT for enhancing multivariate predictive accuracy. However, this study is limited by the exclusion of non-economic external variables such as market sentiment and geopolitical factors. Future research should incorporate additional variables and explore hybrid approaches to achieve more robust gold price forecasting.]]>
Copyrights © 2026
