Stock Market Prediction Model Based on Deep Learning and Enhancement of Interpretabilit

Abstract

This paper's objective is to delve into the utilization of deep learning technology within the realm of stock market forecasting, specifically emphasizing the enhancement of model interpretability. To accomplish this, we employ a deep learning model rooted in the long-term and short-term memory network (LSTM). We proceed to construct three distinct models: the foundational LSTM model, an LSTM model augmented with an attention mechanism, and an LSTM model incorporating an integrated learning strategy. By conducting comparative experiments, we assess the effectiveness of these models in both regression prediction, particularly forecasting the closing price of the following day, and fluctuation prediction. The outcomes of these experiments reveal that integrating an attention mechanism and an integrated learning strategy notably boosts the prediction accuracy of the LSTM model. The attention mechanism, by dynamically assigning weights to various time steps and features, amplifies the model's focus on crucial information, thereby enhancing prediction accuracy. At the same time, the attention mechanism also provides an intuitive explanatory perspective. The integrated learning strategy improves the overall stability and generalization ability by combining the predictions of multiple models. These findings help investors and financial institutions make more informed decisions.