Scholarlink

<p>Stock market prediction has traditionally posed challenges due to the complex and volatile nature of financial markets. This work introduces a deep learning framework integrating Bidirectional Long Short-Term Memory (Bi-LSTM) and Bidirectional Gated Recurrent Units (Bi-GRU) with attention mechanisms to enhance predictive accuracy and interpretability. Unlike conventional methods, these advanced neural architectures can capture long-term dependencies and non-linear patterns in stock data. The addition of attention mechanisms enables the model to focus on critical time points, improving both forecast precision and model clarity. To further enhance transparency, we employ Local Interpretable Model-Agnostic Explanations (LIME), allowing analysts to understand the model’s decision-making process by revealing which factors influence predictions. This approach addresses the need for explainable AI in finance, aiming to bridge the gap between model accuracy and trustworthiness. The results demonstrate that our framework not only outperforms traditional forecasting methods but also offers valuable insights for financial analysts, making it a reliable and interpretable tool for stock market forecasting. This research provides a promising foundation for the future of AI-driven financial decision-making, balancing predictive power with much-needed transparency.</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

Explainable Hybrid Recurrent Models for Stock Price Prediction: Integrating Attention for Transparency

  • 期刊论文

  • 发表日期：

    2026年1月16日

查看全文
  • Apurba Nandi,
  • Sruti Ghorai,
  • Shaoni Banerjee,
  • Arijeet Ghosh,
  • Avik Kumar Das,
  • Sangita Dutta

摘要

Stock market prediction has traditionally posed challenges due to the complex and volatile nature of financial markets. This work introduces a deep learning framework integrating Bidirectional Long Short-Term Memory (Bi-LSTM) and Bidirectional Gated Recurrent Units (Bi-GRU) with attention mechanisms to enhance predictive accuracy and interpretability. Unlike conventional methods, these advanced neural architectures can capture long-term dependencies and non-linear patterns in stock data. The addition of attention mechanisms enables the model to focus on critical time points, improving both forecast precision and model clarity. To further enhance transparency, we employ Local Interpretable Model-Agnostic Explanations (LIME), allowing analysts to understand the model’s decision-making process by revealing which factors influence predictions. This approach addresses the need for explainable AI in finance, aiming to bridge the gap between model accuracy and trustworthiness. The results demonstrate that our framework not only outperforms traditional forecasting methods but also offers valuable insights for financial analysts, making it a reliable and interpretable tool for stock market forecasting. This research provides a promising foundation for the future of AI-driven financial decision-making, balancing predictive power with much-needed transparency.