Forecasting Multi-Stage Time Series of Securities Values Using Various Machine Learning Methods

The graduation thesis titled "Forecasting multistep time series of securities prices using various machine learning methods" covers theoretical aspects, such as a review of past works on related topics, and a practical component - the implementation of these models for forecasting multi-step time series of stock prices. The aim of the study is to examine existing methods for forecasting time series for the value of securities, evaluate their effectiveness, and propose a forecasting model of our own. The object of the study is time series forecasting. The subject of the study is multi-step forecasting of the stock price time series using various machine learning tools and selecting the optimal option. The paper consists of an introduction, three chapters, a conclusion, a list of literature, and an appendix. In the first chapter, a review of existing research in this area is conducted. The second chapter describes the main concepts of machine learning models and their technical description. The main ML models (Linear Regression, Gradient Boosting, SVR), time series models such as ARIMA and ETS, as well as DL models: LSTM, were discussed. The third chapter presents the results of implemented models and the values of hyperparameters achieving the best quality of models. The volume of the work is 60 pages. The paper contains 11 figures, 11 tables, and 2 appendices, and 31 sources of information were used.