Сomparative analysis of machine learning methods for building forecasts

DOI: 10.31673/2409-7292.2024.040013

Abstract

Everything that happens in the world, people try to predict. Previously, forecasting was carried out using observations and calculations performed by people. Today, everything can be predicted using machine learning. Therefore, the topic of machine learning methods is very relevant. This article considers various machine learning methods used for forecasts. The following methods were analyzed: linear regression, random forest, gradient boosting, neural networks, support vector method, k-nearest neighbors method, as well as automatic machine learning. Each of these methods is characterized by selected criteria in order to compare their effectiveness in forecasting. Each method has its own advantages and disadvantages, and it is impossible to determine one universally best, since their effectiveness depends on the specific type of forecasts. To summarize the results of the study, a table was constructed in which the key characteristics of each method are indicated, which allows you to clearly assess their strengths and weaknesses. Special attention is paid to the support vector method, in particular its specificity in forecasting and the possibilities of improving accuracy under certain conditions. The linear regression method, its advantages, disadvantages, and effective application cases are also considered in detail. The article proposes an approach to combining these two methods to create a hybrid approach to forecasting that can improve results in complex problems. The work provides practical recommendations for choosing a machine learning method depending on the forecasting task.

Keywords: machine learning, linear regression, support vector method, forecasting using machine learning.