Stock prediction based on bidirectional gated recurrent unit with convolutional neural network and feature selection.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Financial Forecasting

Background:

• Deep learning algorithms offer superior data pattern recognition and fitting effects compared to traditional machine learning.
• Accurate stock forecasting remains a significant challenge due to market volatility and complex influencing factors.

Purpose of the Study:

• To propose and evaluate a novel hybrid deep learning model for enhanced stock forecasting accuracy.
• To investigate the efficacy of combining Feature Selection, Convolutional Neural Networks, and Bidirectional Gated Recurrent Units for time-series financial data.

Main Methods:

• A hybrid model integrating Feature Selection (FS) for data preprocessing, Convolutional Neural Network (CNN) for local feature extraction, and Bidirectional Gated Recurrent Unit (BGRU) for time-series data processing was developed.
• The proposed FS-CNN-BGRU model was benchmarked against single models (CNN, LSTM, GRU) and hybrid models (CNN-LSTM, CNN-GRU).

Main Results:

• The experimental results indicate that the hybrid FS-CNN-BGRU model significantly outperforms individual models and other hybrid combinations in stock forecasting tasks.
• Feature Selection effectively identified optimal input data, while CNN and BGRU components captured essential local and sequential patterns, respectively.

Conclusions:

• The proposed FS-CNN-BGRU model offers a robust and effective approach for stock market prediction, outperforming existing deep learning architectures.
• This hybrid model provides valuable insights and a strong reference for future research in financial time-series forecasting.