Stock Market Prediction Method Based on Simple Market Feature Data and the Attention Framework

As financial markets become increasingly complex and volatile, traditional statistical and machine learning methods reveal limitations in accurately predicting market trends. The introduction of deep learning techniques, particularly models like Long Short-Term Memory (LSTM) networks and Convolutional Recurrent Neural Networks (CRNN), has significantly enhanced the precision of market forecasts. However, these models often fail to account for the inherent uncertainties of financial markets, leading to poor predictive performance under extreme market conditions. To address this issue, this paper proposes a stock market prediction method that integrates attention mechanisms, bi-directional LSTM, and generative Transformers. By leveraging the attention mechanism and bi-directional LSTM to learn complex patterns within time series data, a model for forecasting stock market trends is constructed. This model is further combined with a generative Transformer to introduce a method for predicting future stock market movements. This paper conducts a performance analysis of the stock market prediction models using Attention as the framework, as well as the combined model (prediction model + generative Transformer), based on selected metrics. The experimental results demonstrate that the stock market prediction model exhibited a good fit across most subjects, with exceptionally strong performance on certain specific cases. The research aims to contribute a novel modeling framework to the field of financial market prediction, analyzing the advantages and disadvantages of this approach and providing insights and guidance for future research.