Machine learning is the main field of artificial intelligence. This contributes to a new stage in the development of the field of information technology, since now the computer is able to switch to self-learning mode without explicit programming. The aim of the study was to find the optimal set of exogenous variables that ensures the best quality of the model in the task of forecasting output volumes. As a result, several methods of constructing new attributes are implemented and the main aspects in the preprocessing of data from this subject area are highlighted.