Research on the effectiveness of ensemble machine learning algorithms in financial forecasting tasks

Abstract

The relevance of the study is driven by the need to improve the accuracy of forecasting enterprises’ financial indicators under conditions of economic instability, growing data volumes, and increasing complexity of their structure. Traditional statistical methods often prove insufficiently effective due to their limited ability to capture nonlinear relationships and the multifactorial nature of financial processes. In this context, machine learning methods—particularly ensemble algorithms - are becoming increasingly important, as they provide enhanced accuracy and robustness of forecasts.

The research problem lies in the absence of a unified approach to selecting and evaluating the effectiveness of ensemble algorithms in forecasting financial indicators of enterprises, as well as insufficient consideration of the specific characteristics of financial data, including their temporal structure, presence of lagged dependencies, and the influence of external factors.

The aim of the study is to examine the effectiveness of ensemble machine learning algorithms and to identify the most appropriate models for forecasting enterprise financial indicators. The research employs machine learning methods, including Random Forest, Gradient Boosting, AdaBoost, Extra Trees, and XGBoost, as well as a baseline linear regression model for comparative analysis. The experiment is conducted using quarterly financial data and evaluated with MAE, RMSE, MAPE, and R² metrics. The methodology includes data preprocessing, formation of lagged features, and splitting the dataset into training and test sets without violating its temporal structure.

The results show that ensemble algorithms provide higher forecasting accuracy compared with baseline models. The best performance was demonstrated by the XGBoost model, which reduced forecast errors by 18–25% across key metrics. It was established that the use of lagged features and a comprehensive approach to factor formation significantly improves model quality.

The key conclusions indicate that ensemble algorithms are an effective tool for financial forecasting and are suitable for implementation in decision support systems, analytical platforms, and FinTech solutions. The obtained results can be used to enhance the validity of managerial decisions and improve enterprises’ adaptability to dynamic economic environments.