Research on Stock Price Prediction Based on LSTM Model

With the increasing complexity of financial markets and the rise of algorithmic trading, deep learning-based stock prediction methods have attracted widespread attention. This study focuses on the daily closing prices of Apple Inc. (AAPL) from 1980 to 2008. Time-series samples were constructed using MinMax normalization and a sliding window approach. The predictive performance of single-layer and double-layer LSTM models (with or without Dropout) was systematically compared against support vector regression (SVR). Experiments were quantitatively evaluated using MSE, RMSE, and MAE metrics. Combined with loss curves and visualization of prediction results, the study found that the single-layer LSTM without Dropout achieved the best performance in terms of RMSE (2.020) and MAE (1.604), while SVR exhibited significantly higher errors. The results indicate that a moderate application of Dropout can effectively enhance model generalization, though it may slightly slow convergence. This study provides intuitive comparisons and optimization suggestions for applying LSTM in financial time-series prediction.