Enhancing Stock Price Prediction Accuracy Through Ensemble Learning Strategies: A Comparative Study

Abstract

This research explores the effectiveness of ensemble learning techniques, including Random Forest, Gradient Boosting, and Stacking, in improving the accuracy and reliability of stock price predictions. Leveraging a dataset of daily trading data for Amicorp Inc. spanning three years, we conducted a comprehensive analysis to investigate the impact of feature engineering, model performance, robustness, and adaptability to varying market conditions. Our findings reveal that feature engineering significantly enhances model performance, with models incorporating additional financial indicators consistently outperforming those without. Among the ensemble methods evaluated, the Random Forest ensemble emerged as the top performer, demonstrating its superiority with the lowest prediction errors. Furthermore, the model displayed robustness in volatile market conditions and resistance to outliers. Market regime analysis highlighted the adaptability of ensemble methods, with consistent performance across bull, bear, and sideways markets. Practical implications were exemplified through a strategic trading strategy based on Random Forest predictions, achieving favorable risk-adjusted returns. These results contribute valuable insights to researchers and practitioners seeking to employ ensemble learning in stock price prediction, underlining its potential for enhancing forecasting accuracy in real-world financial markets.