Scholarlink

<p>The stock market is where investors trade shares of publicly traded companies, with prices influenced by supply and demand, business performance, economic conditions, and investor sentiment. Accurate stock price forecasting is essential for maximizing returns in this volatile environment. Traditional methods struggle to capture the complex depen- dencies in stock price data. This research investigates the performance of Multi-Layered Sequential Long Short-Term Memory (MLS-LSTM) and Gated Recurrent Unit (GRU) neural networks in stock price prediction, using the Adam optimizer for model opti- mization. The study compares the accuracy, training efficiency, and robustness of these models on financial time series data. Additionally, LSTM-based Sentiment Analysis is conducted using data from digital media platforms to gauge public sentiment on stocks. The GRU model achieves a training accuracy of 99.2% and a testing accuracy of 94.8%, outperforming the MLS-LSTM, which has training and testing accuracies of 98.6% and 91.9%, respectively. The GRU’s mean absolute percentage error (MAPE) is 1.39% on the training set and 1.13% on the testing set, better than the MLS-LSTM’s MAPE of 1.81% and 1.38%. The GRU model also has a lower normalized root mean squared error, making it a reliable solution for real-world stock price forecasting.</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

Precision Investing: Combining Gated Recurrent Unit and Sentiment Analysis for Enhanced Stock Market Predictions

  • 期刊论文

  • 发表日期：

    2026年7月28日

查看全文
  • Padmakumari P.,
  • Pranav R. Swaminathan,
  • Harshavardhan S.R,
  • Viswadarsini S.,
  • Jaiaakaash J.S

摘要

The stock market is where investors trade shares of publicly traded companies, with prices influenced by supply and demand, business performance, economic conditions, and investor sentiment. Accurate stock price forecasting is essential for maximizing returns in this volatile environment. Traditional methods struggle to capture the complex depen- dencies in stock price data. This research investigates the performance of Multi-Layered Sequential Long Short-Term Memory (MLS-LSTM) and Gated Recurrent Unit (GRU) neural networks in stock price prediction, using the Adam optimizer for model opti- mization. The study compares the accuracy, training efficiency, and robustness of these models on financial time series data. Additionally, LSTM-based Sentiment Analysis is conducted using data from digital media platforms to gauge public sentiment on stocks. The GRU model achieves a training accuracy of 99.2% and a testing accuracy of 94.8%, outperforming the MLS-LSTM, which has training and testing accuracies of 98.6% and 91.9%, respectively. The GRU’s mean absolute percentage error (MAPE) is 1.39% on the training set and 1.13% on the testing set, better than the MLS-LSTM’s MAPE of 1.81% and 1.38%. The GRU model also has a lower normalized root mean squared error, making it a reliable solution for real-world stock price forecasting.