COMPARATIVE STUDY OF DEEP ENSEMBLE LEARNING MODELS FOR FINANCIAL MARKET PREDICTION
Abstract
This paper presents a comprehensive comparative study of deep ensemble learning models for predicting financial market behavior. With the growing complexity and volatility in financial data, ensemble methods—particularly deep learning-based ensembles—offer significant advantages in terms of prediction accuracy, robustness, and generalization. This study evaluates several architectures including bagging, boosting, stacking, and hybrid ensembles that incorporate deep neural networks such as LSTM, GRU, CNN, and Transformer models. By utilizing historical stock data, the study compares model performance in terms of RMSE, MAPE, and directional accuracy. Results show that deep ensemble learning approaches outperform single deep models and traditional machine learning algorithms across multiple metrics.