Enhancing financial product forecasting accuracy using EMD and feature selection with ensemble models

Abstract

This study examines the impact of Empirical Mode Decomposition (EMD) and Recursive Feature Elimination (RFE) on the prediction of financial product performance employing several ensemble machine learning models, including Random Forest, XGBoost, LightGBM, AdaBoost, CatBoost, Bagging, and ExtraTrees. The research sample comprises ten diverse financial products such as stocks, indices, cryptocurrencies, and commodities. The findings reveal that the combination of EMD and RFE significantly enhances prediction accuracy, with XGBoost delivering the best results. Although all ensemble models benefited from these preprocessing techniques, XGBoost, Random Forest, and LightGBM consistently outperformed the others. This research underscores the critical role of EMD and RFE in improving the predictive capabilities of machine learning models in the dynamic and complex landscape of financial markets, offering valuable insights for practitioners aiming to enhance forecasting accuracy.