<![CDATA[Bitcoin is a non-linear and non-stationary digital asset that has become a highly volatile asset challenging the usual prediction models. In this paper, the authors present a problem-specific Hybrid Ensemble Stacking approach, the proposed approach, which combines the benefits of Long Short-Term Memory (LSTM) in terms of capturing long-term temporal variations with the power of Random Forest (RF) to process complex technical characteristics. The model follows a two-tier structure with a split ratio of 90:10 using BTC/USD historical data of Yahoo Finance and Binance (20102025) to combine the predictions of base learners with the use of a Linear Regression meta-learner. Findings show that pure LSTM has a low RMSE and MAE, but the Hybrid model has the best Mean Absolute Percentage Error (MAPE) of 3.54%. This means that the stacking mechanism will provide a more balanced error percentage, that is, it will enhance stability in forecasting at the phases of price discovery. It is novel in the sense that it uses macro-technical indicators to stabilize predictions in the face of market anomalies as a stacking scheme. These results have real-life implications on developers of financial systems in creating consistent crypto-asset risk management instruments.]]>
Copyrights © 2026
