Predicting Business Bankruptcy in Colombian SMEs: A Machine Learning Approach

Abstract

In this research paper, we address the challenge of predicting business bankruptcy in small and medium-sized enterprises (SMEs) in Colombia. We analyze various financial and non-financial factors that influence the likelihood of bankruptcy and employ machine learning techniques to improve prediction accuracy. We construct a database of 62,500 SMEs for the period 2017–2021 and compare two estimation methods: logistic regression and the eXtreme Gradient Boosting (XGBoost) algorithm. The findings demonstrate that the XGBoost algorithm outperforms in bankruptcy prediction. Key financial variables such as profitability and access to working capital, as well as non-financial variables such as geographic location, are identified as influencing bankruptcy risk. These findings provide valuable insights for stakeholders such as managers, financial intermediaries, and governmental decision-makers in their efforts to support and finance SMEs in Colombia, aiming to reduce bankruptcy rates and promote their economic success.