Scholarlink

<p>Stock price forecasting remains a significant challenge due to market volatility, nonlinear dynamics, and complex interdependencies among financial variables. This study proposes a novel Hybrid Ensemble Model (HEM) that integrates four widely used ensemble learning techniques, namely Bagging, Boosting, Stacking, and Dagging, within a unified meta-learning framework. By aggregating the predictive outputs of individual ensemble models through a linear regression meta-learner, HEM aims to optimize forecasting accuracy while mitigating the limitations inherent in each standalone approach. The model is evaluated on univariate and multivariate representations of META stock data, spanning the period from 2018 to 2024. Performance is assessed using Mean Squared Error (MSE), R-Squared (R<sup>2</sup>), Symmetric Mean Absolute Percentage Error (sMAPE), and Directional Accuracy (DA), alongside rigorous statistical validation tests including the Diebold-Mariano (DM), Modified Diebold-Mariano (M-DM), and Model Confidence Set (MCS) tests. Results demonstrate that HEM consistently outperforms individual ensemble methods in both training and testing phases. In the univariate setting, HEM achieved a significantly lower MSE of 41.92 compared to Bagging (413.26), Boosting (829.16), Stacking (314.44), and Dagging (406.82). In the multivariate context, its MSE approached zero, further affirming its superiority. Additionally, HEM was the only model retained in the MCS, establishing its statistical dominance across forecasting scenarios. The study highlights HEM’s ability to deliver accurate, stable, and generalizable predictions, particularly in dynamic financial environments. Its architecture effectively reduces overfitting, captures directional trends, and enhances model reliability. These findings underscore the value of hybrid ensemble strategies in time series forecasting and position HEM as a scalable and adaptable solution for broader predictive analytics applications in finance and beyond.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Novel Hybrid Ensemble Framework for Stock Price Prediction: Combining Bagging, Boosting, Dagging, and Stacking

  • 期刊论文

  • 发表日期：

    2025年5月7日

查看全文
  • Aqib Gul

摘要

Stock price forecasting remains a significant challenge due to market volatility, nonlinear dynamics, and complex interdependencies among financial variables. This study proposes a novel Hybrid Ensemble Model (HEM) that integrates four widely used ensemble learning techniques, namely Bagging, Boosting, Stacking, and Dagging, within a unified meta-learning framework. By aggregating the predictive outputs of individual ensemble models through a linear regression meta-learner, HEM aims to optimize forecasting accuracy while mitigating the limitations inherent in each standalone approach. The model is evaluated on univariate and multivariate representations of META stock data, spanning the period from 2018 to 2024. Performance is assessed using Mean Squared Error (MSE), R-Squared (R2), Symmetric Mean Absolute Percentage Error (sMAPE), and Directional Accuracy (DA), alongside rigorous statistical validation tests including the Diebold-Mariano (DM), Modified Diebold-Mariano (M-DM), and Model Confidence Set (MCS) tests. Results demonstrate that HEM consistently outperforms individual ensemble methods in both training and testing phases. In the univariate setting, HEM achieved a significantly lower MSE of 41.92 compared to Bagging (413.26), Boosting (829.16), Stacking (314.44), and Dagging (406.82). In the multivariate context, its MSE approached zero, further affirming its superiority. Additionally, HEM was the only model retained in the MCS, establishing its statistical dominance across forecasting scenarios. The study highlights HEM’s ability to deliver accurate, stable, and generalizable predictions, particularly in dynamic financial environments. Its architecture effectively reduces overfitting, captures directional trends, and enhances model reliability. These findings underscore the value of hybrid ensemble strategies in time series forecasting and position HEM as a scalable and adaptable solution for broader predictive analytics applications in finance and beyond.