Application of Machine Learning and Text Data Analysis Methods to Predict Stock Prices in the Russian Market

This final thesis is devoted to the development of an integrated approach to forecasting the value of shares of Russian companies using modern methods of machine learning and natural language processing. The main focus was on finding the best model for obtaining the highest quality forecasts and integrating text data from news sources into the architecture of predictive models. The study analyzed the shares of 215 companies listed on the Moscow Stock Exchange for the period from 2010 to 2024 using more than a million RIA Novosti news articles. Ridge Regression, Random Forest, XGBoost, SERIMAX, and LSTM models with hybrid approaches have been implemented and compared, demonstrating an error of less than 3% in MAPE over a one-day forecasting horizon. A special feature of the work was the creation of an automated web service based on Facet api and Streamlit with cloud data storage integration to provide the results of the work in a convenient way.