Analysis of Stock Price Forecasting Methods based on LSTM Models

Abstract

In the context of today's global economic integration and the dynamic changes in the financial market, the volatility and uncertainty of the stock market have profound impacts on investors, businesses, and policymakers. Traditional technical analysis and fundamental analysis are no longer sufficient to meet the demand for an in-depth understanding of market dynamics. Therefore, exploring more scientific and precise forecasting methods is particularly important. This study focuses on stock price prediction research based on the Long Short-Term Memory (LSTM) machine learning model and optimizes the LSTM model by constructing and analyzing complex networks in the stock market to obtain trait data of the main net inflow complex network. This study constructs a traditional LSTM model to predict historical price data of individual stocks. The model employs a two-layer Long Short-Term Memory (LSTM) network architecture and uses the Adam optimization algorithm along with Mean Squared Error (MSE) as the loss function for training, to measure the model's accuracy in predicting stock prices. Subsequently, this study constructs a complex network of the stock market. Pearson correlation coefficients are obtained from the main force purchase data to build an unweighted correlation network, and key features of the network, such as node degree and betweenness centrality, are extracted. These features, after normalization, form a new set of input features. The study then incorporates the extracted complex network feature data as variables into the traditional LSTM model. The new model structure includes the input layer, hidden layers, output layer, and adapted feature set. During the model training process, batch normalization and Dropout techniques are used to enhance the model's generalization ability. Experimental results show that the improved model has significantly reduced prediction errors in multiple time windows. The innovation of this study lies in combining complex network analysis data with the LSTM model, focusing on the internal structural characteristics of the stock market while emphasizing the dynamics and real-time nature of market forecasting. Additionally, Independent Component Analysis (ICA) is used for preprocessing market data, reducing noise interference, and improving the accuracy of the prediction model.