Research on Gold Price Prediction Based on VMD-SVM

Gold prices are influenced by multiple factors, including market supply and demand, global economic conditions, and policy changes, exhibiting high nonlinearity and non-stationarity. Accurate prediction of gold price trends is crucial for financial investment and risk management. Existing deep learning and traditional machine learning methods often face challenges such as overfitting, large errors, and insufficient robustness in gold price forecasting. To address these issues, this paper proposes the VMD-SVM model. First, variational mode decomposition (VMD) is applied to decompose the time series, reducing noise interference. Then, support vector machines (SVMs) are trained separately to predict each decomposed subseries, and the final result is reconstructed. Experimental results show that VMD-SVM achieves the lowest mean absolute error and the highest coefficient of determination on the test set, significantly outperforming LSTM, random forest, and SVM. These findings validate the effectiveness of VMD in enhancing SVM's prediction performance. The study demonstrates that VMD-SVM offers superior prediction accuracy and generalization capability, providing a novel approach to financial market time series modeling and extending its applicability to energy price forecasting and macroeconomic analysis.