<![CDATA[This study presents a comparative analysis of Long Short-Term Memory (LSTM) and Support Vector Regression (SVR) performance in predicting gold futures prices (XAU/USD). Using historical data from October 2020 to October 2025 (1,289 data points) covering the COVID-19 pandemic and post-pandemic periods, both models were evaluated under two scenarios: default parameters and optimized hyperparameters. Results showed significant performance differences without tuning, where LSTM achieved R² of 0.9874 while SVR produced negative R² of -6.3571. However, after hyperparameter optimization using 27 configurations for each model, both methods demonstrated excellent and comparable performance. Optimized SVR achieved R² of 0.9887, MAPE of 0.9402%, RMSE of $39.21, and MAE of $29.65, while optimized LSTM obtained R² of 0.9895, MAPE of 0.8996%, RMSE of $38.18, and MAE of $28.32. Paired t-test revealed no statistically significant difference between the two methods after tuning (p=0.0978 > 0.05), with both models exhibiting excellent generalization capabilities (R² gap < 0.005). These findings demonstrate that hyperparameter tuning is more critical than algorithm selection in achieving high prediction accuracy, suggesting that optimized SVR can serve as an efficient alternative to LSTM for applications with limited computational resources]]>
Copyrights © 2026
