Scholarlink

<p>Predicting stock prices has been a difficult challenge because of the complexity, uncertainty, unpredictability, and disorder of the stock market. Its accuracy and profitability could be greatly increased by using deep learning models, especially when combined with a large amount of high-quality data, cautious validation, and interpretation. This paper examines the deep learning models that are feasible for price index data. Here, a novel hybrid deep learning model, which employs a multilayer perceptron (MLP), a bidirectional long-short term memory neural network (BiLSTM), and a convolutional neural network (CNN), is proposed, integrated with a range of technical analysis indicators for predicting the closing prices of the National Stock Exchange FIFTY (NIFTY 50) index. In order to verify the improved performance of our proposed hybrid model on the pricing task, several competing models were built. We choose the principal component analysis (PCA) technique and the Pearson correlation coefficient method to select the most impactful technical indicators and reduce the feature dimension, to improve the stability and interpretability of our pricing models. As a result, we utilized a set of 18 predictors based on technical indicators and fundamental (historical) market data to capture the fluctuations of the stock market. The proposed model has been compared to the other baseline models using an extensive model evaluation process. Additionally, we carried out robustness check and statistical analysis to test the significance of the proposed model’s performance to other baseline models, ensuring reliability and robustness. After rigorous experimentation, the proposed model has consistently proven its exceptional predictive performance, achieving the lowest error values (RMSE of 0.019623 and MedAE of 0.006510). With the help of the Pearson correlation coefficient and the PCA technique, we demonstrated that the performance given by the proposed model could be attributed to its ability to capture time-series data and its focus on input features.</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

A Hybrid MLP-Based Framework for Stock Price Prediction Using Technical Indicators

  • 期刊论文

  • 发表日期：

    2026年4月8日

查看全文
  • Shubhangi Porwal,
  • Namita Srivastava,
  • Manoj Jha

摘要

Predicting stock prices has been a difficult challenge because of the complexity, uncertainty, unpredictability, and disorder of the stock market. Its accuracy and profitability could be greatly increased by using deep learning models, especially when combined with a large amount of high-quality data, cautious validation, and interpretation. This paper examines the deep learning models that are feasible for price index data. Here, a novel hybrid deep learning model, which employs a multilayer perceptron (MLP), a bidirectional long-short term memory neural network (BiLSTM), and a convolutional neural network (CNN), is proposed, integrated with a range of technical analysis indicators for predicting the closing prices of the National Stock Exchange FIFTY (NIFTY 50) index. In order to verify the improved performance of our proposed hybrid model on the pricing task, several competing models were built. We choose the principal component analysis (PCA) technique and the Pearson correlation coefficient method to select the most impactful technical indicators and reduce the feature dimension, to improve the stability and interpretability of our pricing models. As a result, we utilized a set of 18 predictors based on technical indicators and fundamental (historical) market data to capture the fluctuations of the stock market. The proposed model has been compared to the other baseline models using an extensive model evaluation process. Additionally, we carried out robustness check and statistical analysis to test the significance of the proposed model’s performance to other baseline models, ensuring reliability and robustness. After rigorous experimentation, the proposed model has consistently proven its exceptional predictive performance, achieving the lowest error values (RMSE of 0.019623 and MedAE of 0.006510). With the help of the Pearson correlation coefficient and the PCA technique, we demonstrated that the performance given by the proposed model could be attributed to its ability to capture time-series data and its focus on input features.