Scholarlink

<p>Stock market price prediction is important in the financial sectors for better investment decisions and efficient risk management against market volatility. Stock market data are inherently noisy and volatile. Quality and quantity of input data control the stability of any deep learning (DL) prediction model and hence, denoising is quite essential. Herein, we propose a new framework, namely Bi-GRU-SG, including Bidirectional Gated-Recurrent-Unit (Bi-GRU) and Savitzky-Golay (SG) filter to obtain enhanced stock market prediction accuracy. The contributions of the paper are (i) utilisation of SG filtering to eliminate short-term stock fluctuations and to preserve long-term relevant stock patterns that make the stock market patterns more evident for the Bi-GRU model (ii) integration of Bi-GRU model to capture the complete temporal dependencies in the stock data. Unlike conventional smoothing processes, SG filtering is computationally cheap and retains the important trends such as turning points, peak values, and gradient information of the stock. Whereas, the Bi-GRU model efficiently preserves rich temporal features of the stock data along both forward and backwards directions, enabling a robust and accurate stock prediction. Performance of the proposed model is validated on the CSI-300 daily dataset and is compared with some of the baseline prediction models. The proposed model, integrating a trend-preserving smoothing process along with a bi-directional sequence model, provides enhanced prediction precision and stability.</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

An efficient framework for accurate stock market price prediction using bidirectional GRU and Savitzky-Golay filter

  • 期刊论文

  • 发表日期：

    2025年11月11日

查看全文
  • Nirupama Parida,
  • Bunil Kumar Balabantaray,
  • Rajashree Nayak

摘要

Stock market price prediction is important in the financial sectors for better investment decisions and efficient risk management against market volatility. Stock market data are inherently noisy and volatile. Quality and quantity of input data control the stability of any deep learning (DL) prediction model and hence, denoising is quite essential. Herein, we propose a new framework, namely Bi-GRU-SG, including Bidirectional Gated-Recurrent-Unit (Bi-GRU) and Savitzky-Golay (SG) filter to obtain enhanced stock market prediction accuracy. The contributions of the paper are (i) utilisation of SG filtering to eliminate short-term stock fluctuations and to preserve long-term relevant stock patterns that make the stock market patterns more evident for the Bi-GRU model (ii) integration of Bi-GRU model to capture the complete temporal dependencies in the stock data. Unlike conventional smoothing processes, SG filtering is computationally cheap and retains the important trends such as turning points, peak values, and gradient information of the stock. Whereas, the Bi-GRU model efficiently preserves rich temporal features of the stock data along both forward and backwards directions, enabling a robust and accurate stock prediction. Performance of the proposed model is validated on the CSI-300 daily dataset and is compared with some of the baseline prediction models. The proposed model, integrating a trend-preserving smoothing process along with a bi-directional sequence model, provides enhanced prediction precision and stability.