Hybrid decomposition and deep learning approach for data-driven FOREX forecasting optimization

Predicting the highly volatile foreign exchange (FOREX) market is a challenging task influenced by economic, geopolitical, and psychological factors. Sudden market fluctuations and unexpected events further complicate this endeavor. The present study introduces a novel hybrid regression approach for FOREX rate prediction, integrating empirical mode decomposition, a stacked long short-term memory deep learning model, and the particle swarm optimization metaheuristic into a unified framework. The proposed method is evaluated on three major currency pairs, namely EUR/USD, USD/CHF, and EUR/CHF, and benchmarked against its standalone components as well as state-of-the-art machine learning models, including XGBoost and support vector regression. Additionally, its effectiveness as a signal provider is tested through a simulated trading environment focused on the EUR/USD pair. The results demonstrate that the proposed approach outperforms competing models, achieving high profitability while minimizing risk.