Stock Price Prediction Model Based on Blending Model Improved with Sentiment Factors and Double Q-learning

Stock market prediction is of paramount importance for economic decision-makers and investors. Traditional approaches often rely on single indicators or sentiment analysis techniques, which may limit their ability to capture the complexity of market dynamics. In this research, we propose a novel multi-dimensional approach that integrates investor sentiment indicators, market sentiment indicators, and emergency event indicators to construct a comprehensive stock market prediction model. Our methodology leverages a blending learning strategy, employing various base classifiers including cnn-lstm, LightGBM, Random Forest, and logistic regression etc. This diverse set of models allow us to evaluate sentiment indicators from different perspectives, enhancing the robustness and accuracy of predictions. To address the limitations of fixed base classifier numbers in blending models, we introduce the Double Q-learning method from reinforcement learning. This dynamic weight adjustment mechanism optimizes the number and weights of base classifiers and based on their predictive effectiveness, improving the model's adaptability and prediction accuracy. Our research not only demonstrates innovation in stock market prediction but also provides a flexible and effective methodology for complex financial market analysis.