Development and Research of a Modified Gradient Boosting Method Effectiveness to Solve Applied Problems of Time-Series Forecasting

Abstract

The article considers the problem of increasing the accuracy of the gradient boosting method when forecasting timeseries (including demand for equipment or other goods). The aim of the study is to develop a modified gradient boosting method of decision trees for this forecasting. The relevance of the work is due to the fact that the subject area under consideration is characterized by the presence of many different noises and anomalies in datasets. The use of machine learning methods in forecasting such data leads to a decrease in the accuracy of forecasts and the occurrence of the overfitting effect. Despite the tunings of the existing parameters in existing gradient boosting algorithms, this problem remains and leads to the need for additional improvements. The object of the study is time-series. The subject of the study is the indicators (or metrics) of the effectiveness of machine learning methods used to forecast these time-series. As a result, the most suitable methods for reducing the effect of overfitting and the influence of anomalies were considered and identified, which formed the basis for the development of a modified method of gradient boosting trees. Finally, conclusions are drawn.