<![CDATA[This research addresses a significant technical gap in Bitcoin price forecasting, specifically the extrapolation failure encountered by Deep Learning and tree-based models when prices reach unprecedented record highs (All-Time Highs). Conventional models often struggle to predict values outside their historical training range, leading to significant inaccuracies during price discovery phases. To solve this issue, a Stacking Hybrid framework is proposed, integrating non-linear components (Random Forest and Gradient Boosting) to capture volatility with a linear component (Ridge Regression) to maintain growth trends. The study utilizes 5,406 daily Bitcoin price observations from 2010 to 2025 and extracts seven key technical features, including the Relative Strength Index (RSI) and Simple Moving Averages (SMA), using the Patterned Dataset technique. Experimental results demonstrate that while single tree-based models fail during the aggressive market phases of 2025 due to their rigid data partitioning, the proposed Stacking Hybrid model achieves superior performance with a Mean Absolute Error (MAE) of $936 and a correlation coefficient ( ) of 0.99. The integration of Ridge Regression as a meta-learner successfully enables price extrapolation beyond historical limits. This framework contributes a robust methodology to the field of Computer Science for handling highly volatile time-series data in digital asset markets.]]>
Copyrights © 2026
