Research on Gold Price Prediction Model Based on CEEMDAN-XGBoost

Abstract

This study proposes a hybrid model that combines Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) and Extreme Gradient Boosting (XGBoost) for predicting the time series of gold prices. Addressing the performance bottlenecks of traditional XGBoost when dealing with nonlinear and non-stationary signals, this research first applies CEEMDAN to decompose historical gold price data into multiscale signals, extracting Intrinsic Mode Functions (IMFs) of different frequency components. These IMFs are then used as input features for training and prediction with the XGBoost model. Experimental results show that the CEEMDAN-XGBoost model achieves high prediction accuracy on both the training and testing sets, outperforming the standalone XGBoost model in terms of generalization ability and prediction performance. This study not only provides an effective modeling approach for gold price prediction but also offers new insights for handling nonlinear and non-stationary features in other financial time series data.