Stock Closing Price Forecasting Using LSTM, Sentiment Analysis, Kalman Filter

The article explores the problem of predicting the closing prices of stocks using an innovative approach that combines methods of machine learning, sentiment analysis and noise filtering. The focus is on the development and testing of a comprehensive model that integrates recurrent neural networks (LSTM) for time series analysis, the GPT-3 language model for processing and analyzing textual data from news and a Kalman filter to improve prediction accuracy by smoothing out the influence of noise on inputs data. The importance of taking into account information from the mass media and social networks to improve the predictive ability of the model is also considered. The purpose of the study is to demonstrate how the integration of various methods and technologies can improve the effectiveness of financial performance forecasting using the example of stock closing prices. The paper presents the results of an experimental evaluation of the proposed approach, showing its advantages over traditional analysis methods. Particular attention is paid to the analysis and discussion of the results of using the proposed model, including the assessment of its effectiveness using both traditional metrics (for example, RMSE) and a specially developed Win Ratio (WR) metric, which allows one to evaluate the model's ability to predict the direction of price changes. The results of the study confirm that the use of a multifactor approach allows one to achieve higher accuracy of forecasts, which can be useful for investors and specialists in the field of financial analysis. The article concludes with a discussion of future research prospects, including the possibility of improving the model through deeper analysis of textual information and the integration of additional data, such as global economic indicators and information verified using blockchain technology. The proposed approach and the results obtained demonstrate significant potential for the development of forecasting methods in financial markets and can contribute to making more informed investment decisions.