Stock Price Prediction Using a Stacked Heterogeneous Ensemble

Abstract

Forecasting stock price ranges remains a significant challenge because of the non-linear nature of financial data. This study proposes and evaluates a stacking ensemble model for range-based volatility forecasting, using open, high, low, and close (OHLC) prices. The model integrates a diverse, heterogeneous set of base learners, such as statistical (ARIMA), machine learning (Random Forest), and deep learning (LSTM, GRU, Transformer) models, with an XGBoost meta-learner. Applied to several major financial indices and a single stock, the proposed framework demonstrates high predictive accuracy, achieving (Formula presented.) scores between 0.9735 and 0.9905. These results highlight the efficacy of a multi-faceted stacking approach in navigating the complexities of financial forecasting.