A Stock Price Prediction Model Based on VMD-SSA-LSTM

The equity market is a dynamic non-linear system full of various types of noise, making accurate prediction a challenging task. To tackle the issues of low prediction precision in traditional single-stock prediction models and the complexity of training traditional neural networks, this paper constructs a VMD-SSA-LSTM model based on stock data of Advanced Micro Devices (AMD). This model is a stock closing price forecasting model built on LSTM and optimized by using VMD and Sparrow Search SSA. The proposed approach first uses VMD to decompose the stock-related data into multiple IMFs to obtain components with different frequency elements. Then, the SSA algorithm is applied to extract frequency domain features from the mode functions in LSTM, capturing key structures and periodicity in the time series. Finally, individual LSTM prediction models are established for each mode component, and the predicted values of all components are aggregated to form the total predicted closing price. Experimental results demonstrate that the VMD-SSA-LSTM combination model, compared to other models, can effectively handle complex time series data, significantly improving prediction accuracy and validating the effectiveness of the method.