STAGE framework: A stock dynamic anomaly detection and trend prediction model based on graph attention network and...

Area of Science:

• Financial technology
• Machine learning in finance
• Data science

Background:

• Financial markets are increasingly complex, demanding advanced tools for risk management.
• Existing stock prediction and anomaly detection methods struggle with intricate inter-stock relationships and data anomalies.

Purpose of the Study:

• To introduce the STAGE framework for enhanced stock prediction and robust anomaly detection.
• To address the limitations of current financial data analysis techniques.

Main Methods:

• Integration of Graph Attention Network (GAT), Variational Autoencoder (VAE), and Sparse Spatiotemporal Convolutional Network (STCN) into the STAGE framework.
• Utilizing a combination of deep learning architectures for spatiotemporal analysis and pattern recognition.

Main Results:

• The STAGE framework achieved 85% accuracy in stock prediction after 20 epochs, surpassing other models by 10-20%.
• Achieved 95% accuracy in anomaly detection, demonstrating fast convergence and stability.
• The framework proved effective in handling complex financial market dynamics.

Conclusions:

• The STAGE framework provides an innovative and effective solution for stock prediction and anomaly detection in complex financial markets.
• The integrated approach enhances accuracy and robustness compared to existing methods.