Modeling Financial Distress Prediction Using XGBoost Ensemble Learning Algorithm

Articles in Press

Document Type : Original Article

Authors

• Hamed Alizadeh Birjandi ¹
• Karim Nakhaie ²
• Arzoo Khosravani ³

¹ PhD student in Accounting, Semnan branch, Islamic Azad University, Semnan, Iran.

² Assistant Professor of Accounting Department, Birjand Branch, Islamic Azad University

³ Assistant Professor of Accounting Department, Semnan Branch, Islamic Azad University.

https://doi.org/10.22067/mfe.2025.92274.1513

Abstract

The aim of the current research is to modeling of financial distress prediction using XGBoost ensemble learning algorithm in listed companies on Tehran Stock Exchange. The statistical population of the present research is the listed companies to the Tehran Stock Exchange during the period 2012 to 2023 and the screening sampling method, in this regard, 1800 firm-year (150 firms for 12 years), observations collected from the annual financial reports of the case have been tested. In this research, independent variables according to the theoretical foundations and empirical background of internal and external research include 52 variables in two categories of accounting variables (profitability indicators, obligations fulfillment indicators, activity indicators, cash flow indicators and growth sustainability indicators) and Non-accounting variables (cash flow composition indices, corporate governance indices, macroeconomic indices, management ability index, audit index and competitiveness index) were determined. In order to identify the important variables for developing the model, the average comparison test of two sample was used, and according to the results, 40 variables out of 52 variables were selected as the final variables for developing the model. The results show that the overall accuracy of XGBoost ensemble learning algorithm and logit regression is 97.8% and 92.1%, respectively, which indicates that compared to logit regression, XGBoost algorithm performs better in predicting companies with Has financial distress. In other words, the results show the efficiency of XGBoost algorithm compared to logit regression. Therefore, the XGBoost algorithm provides the most efficient model for predicting financial distress in listed companies on the Tehran Stock Exchange.

Keywords

• Financial Distress
• Ensemble Learning
• XGBoost Algorithm
• Logit Regression

Main Subjects

• Audit accounting