Scholarlink

<p>Accurate stock price forecasting remains a formidable challenge in both local and global financial markets due to the high volatility and non-linear nature of stock data. Traditional forecasting models, including Long Short-Term Memory (LSTM) networks, have been widely employed; however, they often fall in capturing complex temporal patterns and delivering high predictive accuracy. This study proposes a novel hybrid forecasting framework that integrates Principal Component Analysis (PCA), for dimensionality reduction, with an enhanced LSTM model incorporating Geometric Mean-Based Layer Normalization (GLN), referred to as PCA-GLN-LSTM. PCA is utilized to extract the most significant features from a set of technical indicators, thus reducing noise and computational complexity. The LSTM network, augmented with GLN, facilitates stable and efficient learning by preserving scale-sensitive patterns in time series data. Experimental evaluations conducted on stock data from Life Insurance Corporation of India and publicly available Yahoo Finance data sets demonstrate that the proposed PCA-GLN-LSTM model consistently outperforms baseline approaches. Model performance is rigorously assessed using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE). In addition, the statistical significance of forecasting improvements is validated using the Diebold-Mariano (DM) test at specified significance level. The results substantiate the efficacy of the proposed method in enhancing predictive accuracy for stock price forecasting tasks.</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

Optimizing the accuracy of stock price prediction: accessing the prospects of LSTM with geometric mean layer normalization

  • 期刊论文

  • 发表日期：

    2026年5月8日

查看全文
  • Yuvaraja Boddu,
  • Manimaran A

摘要

Accurate stock price forecasting remains a formidable challenge in both local and global financial markets due to the high volatility and non-linear nature of stock data. Traditional forecasting models, including Long Short-Term Memory (LSTM) networks, have been widely employed; however, they often fall in capturing complex temporal patterns and delivering high predictive accuracy. This study proposes a novel hybrid forecasting framework that integrates Principal Component Analysis (PCA), for dimensionality reduction, with an enhanced LSTM model incorporating Geometric Mean-Based Layer Normalization (GLN), referred to as PCA-GLN-LSTM. PCA is utilized to extract the most significant features from a set of technical indicators, thus reducing noise and computational complexity. The LSTM network, augmented with GLN, facilitates stable and efficient learning by preserving scale-sensitive patterns in time series data. Experimental evaluations conducted on stock data from Life Insurance Corporation of India and publicly available Yahoo Finance data sets demonstrate that the proposed PCA-GLN-LSTM model consistently outperforms baseline approaches. Model performance is rigorously assessed using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE). In addition, the statistical significance of forecasting improvements is validated using the Diebold-Mariano (DM) test at specified significance level. The results substantiate the efficacy of the proposed method in enhancing predictive accuracy for stock price forecasting tasks.