Research on Stock Price Prediction based on the GWO-LSTM-ATT Model

To address the challenge of deep learning models becoming trapped in local optima and to enhance prediction accuracy in stock price forecasting, this study proposes a stock prediction model that utilizes the Grey Wolf Optimization (GWO) algorithm to optimize the Long Short-Term Memory with Attention Mechanism (LSTM-ATT). The GWO algorithm is employed to fine-tune the hyperparameters of the LSTM-ATT model, thereby identifying the optimal hyperparameter configuration necessary for constructing a high-accuracy stock prediction model. Experimental results indicate that for 5-step, 10-step, and 20-step predictions, the Mean Absolute Error (MAE) of the GWO-LSTM-ATT model is reduced by 61.33%, 77.06%, and 89.28%, respectively, in comparison to the LSTM-ATT model. The GWO-LSTM-ATT model not only effectively mitigates the issue of local optima but also offers a novel perspective and a robust tool for enhancing the accuracy and stability of stock price predictions.