A dual-path convolutional neural network combined with an attention-based bidirectional long short-term memory network for stock price prediction

Guiyan Zhao ¹, Yunfei Cheng ², Jianhui Yang ¹

⁰Qiaoxing School of Economics and Management, Fujian Polytechnic Normal University, Fuzhou, China.

Plos One +

|

April 22, 2025

View abstract on PubMed

Summary

Accurate stock price prediction is challenging due to complex data. The new DCA-BiLSTM model, using dual-path convolutional neural networks with attention and bidirectional long short-term memory networks, significantly improves forecasting accuracy.

Area Of Science

* Computational Finance
* Machine Learning
* Time Series Analysis

Background

* Stock price data exhibits nonlinearity, non-stationarity, and complex spatiotemporal patterns, posing significant prediction challenges.
* Traditional forecasting methods often struggle to capture the intricate dependencies within financial time-series data.

Purpose Of The Study

* To develop and validate a novel deep learning model for enhanced stock price forecasting.
* To address the limitations of existing methods in handling complex financial time-series characteristics.

Main Methods

* Proposed DCA-BiLSTM model combining dual-path convolutional neural networks with attention (DCA) and bidirectional long short-term memory networks (BiLSTM).
* Utilized wavelet packet decomposition for extracting high- and low-frequency features.
* DCA module for robust deep feature extraction, followed by BiLSTM for modeling bidirectional dependencies.

Main Results

* The DCA-BiLSTM model demonstrated superior performance across multiple stock datasets (Apple, Google, Tesla) and the Nasdaq index.
* Achieved high R-squared values: 0.9507 (Apple), 0.9595 (Google), 0.9077 (Tesla), and 0.9594 (Nasdaq).
* Showcased significant reductions in error metrics compared to traditional forecasting approaches.

Conclusions

* The DCA-BiLSTM model offers robust and accurate stock price forecasting capabilities.
* The findings provide reliable insights for financial market analysis and investment strategies.
* The model's effectiveness highlights the potential of advanced deep learning techniques in financial time-series prediction.