Scholarlink

Trading professionals can make well-informed decisions about what to purchase or sell in order to maximize short-term gains by forecasting stock prices for the next day. This research study focuses on exploring the compatibility of ensemble learning techniques through stacking to predict next-day stock prices. The models involved—Random Forest, Extra Trees, AdaBoost, and Gradient Boosting, were paired two at a time, and their predictions were used as inputs to a Multi-Layer Perceptron (MLP) Regressor, which served as the meta-learner. The results revealed that the combination of Extra Trees Regressor and Gradient Boosting outperformed the individual base models, due to their complementary strengths and ability to capture non-linear relationships effectively. However, other model combinations showed only average performance. This outcome was attributed to overlapping model strengths, leading to increase in error and overfitting. The findings highlight the importance of thoughtful model selection in ensemble methods and suggest that not all combinations are equally beneficial. Understanding the compatibility of different models is crucial to improving performance in ensemble learning.

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

Prediction of Next-Day Stock Price Using Stacked Ensemble Learning Techniques—An Exploration of Model Compatibility

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Amal Joseph,
  • Bansi Pambhar,
  • Allen George,
  • Sujatha Arun Kokatnoor,
  • Sandeep Kumar

摘要

Trading professionals can make well-informed decisions about what to purchase or sell in order to maximize short-term gains by forecasting stock prices for the next day. This research study focuses on exploring the compatibility of ensemble learning techniques through stacking to predict next-day stock prices. The models involved—Random Forest, Extra Trees, AdaBoost, and Gradient Boosting, were paired two at a time, and their predictions were used as inputs to a Multi-Layer Perceptron (MLP) Regressor, which served as the meta-learner. The results revealed that the combination of Extra Trees Regressor and Gradient Boosting outperformed the individual base models, due to their complementary strengths and ability to capture non-linear relationships effectively. However, other model combinations showed only average performance. This outcome was attributed to overlapping model strengths, leading to increase in error and overfitting. The findings highlight the importance of thoughtful model selection in ensemble methods and suggest that not all combinations are equally beneficial. Understanding the compatibility of different models is crucial to improving performance in ensemble learning.