A Two-Stage Deep Fusion Integration Framework Based on Feature Fusion and Residual Correction for Gold Price Forecasting

Abstract

Given the far-reaching impact of the gold price on global financial markets, accurately predicting the gold price has become essential, with machine learning methods emerging as a prominent tool to tackle this challenge. Nonetheless, traditional single prediction models usually suffer from limited predictive performance and fail to capture complex variability of market behavior. Aiming to solve these limitations, an innovative two-stage hybrid deep integration framework that combines feature extraction and residual correction techniques is proposed with a view to predicting the gold price more accurately. The prediction effectiveness is enhanced by employing a variational modal decomposition to cluster time series data into three classes. The first stage employs variational mode decomposition to categorize time series data, improving computational efficiency and initial prediction accuracy. The second stage refines these predictions through a novel residual correction process, leveraging back propagation, long and short-term memory, and convolutional neural networks. In addition, through the in-depth analysis and processing of residuals, it is demonstrated that starvation of our method further improves the credibility of the prediction results, and effectively predicts the price movements of the four major gold markets. This approach not only provides a remarkably valuable perspective for policy makers, investors, and trading firms in the gold market, but also deals with the shortcomings of a single model in the face of complex market dynamics, and lays the foundation for the development of even more powerful forecasting models in the future.

Keywords

• Feature fusion
• integration model
• price forecast
• residual correction