An enhanced Transformer framework with incremental learning for online stock price prediction

Area of Science:

• Artificial Intelligence
• Machine Learning
• Financial Forecasting

Background:

• Existing stock prediction models struggle with real-time data streams, showing poor scalability and performance degradation.
• Dynamic changes in data distribution and price non-stationarity pose significant challenges for accurate forecasting.

Purpose of the Study:

• To propose an incremental learning-based enhanced Transformer framework (IL-ETransformer) for online stock price prediction.
• To enhance model adaptability to dynamic financial markets and improve prediction accuracy.

Main Methods:

• Leveraging a multi-head self-attention mechanism to analyze temporal dependencies between stock prices and feature factors.
• Employing a continual normalization mechanism for data stream stabilization.
• Utilizing a time series elastic weight consolidation (TSEWC) algorithm for efficient incremental training.

Main Results:

• The IL-ETransformer effectively captures temporal information and exploits correlations among multi-dimensional features.
• Experimental results on five datasets show significant improvements in stock price prediction accuracy.
• The method demonstrates robust performance with non-stationary and frequently changing financial market data.

Conclusions:

• The proposed IL-ETransformer framework offers a scalable and adaptive solution for online stock price prediction.
• It overcomes limitations of traditional models in handling real-time, dynamic financial data.
• The approach enhances predictive accuracy by integrating incremental learning with advanced Transformer architecture.