Advanced Stock Price Prediction Using LSTM and Informer Models

Abstract

As a pivotal component of the global economic system, the stock market is subject to a multitude of influences, including the macroeconomic environment, market sentiment, and policy changes. Consequently, the ability to forecast stock prices is of paramount importance. Conventional time series forecasting techniques, such asARIMA and GARCH, are ill-equipped to handle complex nonlinear relationships. In contrast, recurrent neural networks, particularly Long Short-Term Memory (LSTM) networks, are particularly adept at handling time-dependent data. In light of recent advances in machine learning and deep learning, this study aims to assess and compare the efficacy of LSTM neural networks and Informer models in stock price forecasting. The objectives of this research are twofold: first, to compare the prediction accuracy using metrics such as Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and R²; and second, to explore fusion strategies to enhance overall prediction performance and computational efficiency. The methodology includes the following steps: data collection and preprocessing, model construction, feature engineering, and model training and evaluation. This study presents a systematic comparison of the effectiveness of LSTM and Informer models in stock price prediction. The findings indicate that a fusion strategy combining the advantages of both models is expected to enhance prediction accuracy and computational efficiency.