Scholarlink

<p>This research paper addresses critical challenges in financial forecasting, focusing on enhancing prediction accuracy and reliability. A primary emphasis is placed on optimizing window length in financial time series analysis, as traditional fixed lengths often fail to capture the dynamic market conditions. By dynamically adjusting the window length based on market volatility, the analysis better reflects current trends and improves forecasting outcomes. Additionally, the integration of financial news sentiment with technical indicators is explored, recognizing that qualitative sentiment data can serve as a leading indicator of market movements. The study introduces a novel architecture, the Convolutional Transformer (CNN Transformer), which employs a multi-head convolutional self-attention layer to capture sequential dependencies and a decoder for interpreting long-term relationships. The innovative window length optimization algorithm enhances Mean Absolute Error (MAE) by 19.5% and Root Mean Square Error (RMSE) by 4%. Incorporating news data further boosts MAE by 19% and RMSE by 11.5%. When applied to the Apple stock price dataset, this model significantly outperforms existing KNN models, achieving improvements of 58.55% in RMSE and 65.77% in MAE. The framework’s adaptability to volatile markets makes it suitable for real-time trading systems.</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

Enhancing Financial Forecasting with Dynamic Window Optimization, Sentiment Analysis, and Convolutional Transformers

  • 期刊论文

  • 发表日期：

    2026年2月25日

查看全文
  • Dilip Kumar Sharma,
  • Ravi Prakash Varshney,
  • Anuradha Pillai,
  • Neha Agarwal,
  • Saurabh Agarwal

摘要

This research paper addresses critical challenges in financial forecasting, focusing on enhancing prediction accuracy and reliability. A primary emphasis is placed on optimizing window length in financial time series analysis, as traditional fixed lengths often fail to capture the dynamic market conditions. By dynamically adjusting the window length based on market volatility, the analysis better reflects current trends and improves forecasting outcomes. Additionally, the integration of financial news sentiment with technical indicators is explored, recognizing that qualitative sentiment data can serve as a leading indicator of market movements. The study introduces a novel architecture, the Convolutional Transformer (CNN Transformer), which employs a multi-head convolutional self-attention layer to capture sequential dependencies and a decoder for interpreting long-term relationships. The innovative window length optimization algorithm enhances Mean Absolute Error (MAE) by 19.5% and Root Mean Square Error (RMSE) by 4%. Incorporating news data further boosts MAE by 19% and RMSE by 11.5%. When applied to the Apple stock price dataset, this model significantly outperforms existing KNN models, achieving improvements of 58.55% in RMSE and 65.77% in MAE. The framework’s adaptability to volatile markets makes it suitable for real-time trading systems.