USING MACHINE LEARNING METHODS FOR TIME SERIES FORECASTING: CURRENT APPROACHES AND PERSPECTIVES

The article discusses the use of machine learning techniques for time series forecasting, which is an important task in various fields of life, such as economics, finance, meteorology and business analytics. Traditional statistical methods, including ARIMA models, have demonstrated limited applicability when dealing with nonlinear temporal data. In this regard, modern approaches based on machine learning have gained an actual importance. The main focus of the article is made on such methods as artificial neural networks, nearest neighbor regression and support vector machines, which demonstrate high accuracy and stability degree when working with massive datasets. The purpose of the study is to analyze existing approaches to time series forecasting using machine learning, identify advantages and disadvantages of different algorithms, and discuss methods for single-step and multi-step forecasting. The article emphasizes importance of local learning as a flexible tool for dealing with non-stationary data and also offers prospects for further development in this area as well.