Gerenciamento de risco de crédito por meio da utilização de aprendizado de máquina: o caso do Banco BS2

Credit risk has played a central role in several global financial crises over the past three decades. An increasingly complex and interconnected financial landscape makes risk management essential for the stability and growth of financial institutions. This case study aims to analyze the use of machine learning specifically, the Gradient Boosting Decision Tree (GBDT) algorithm in a predictive model that combines significant financial and non-financial variables and incorporates credit bureau inquiries into Banco BS2’s credit risk management process. The goal is to achieve greater accuracy in decision-making and improvements in risk mitigation. The F1 metric, employed as a measure of the model’s precision, shows a superior value of 0.77 when compared with the model used by Serasa. Since 2022. the continuous monitoring capability offered by this predictive model has provided BS2 Bankwith a real-time view of the financial health of its customer base, thereby facilitating the implementation of more assertive policies. Furthermore, the default rate among Banco BS2’s corporate clients, as recorded by BCB-CADOC (2024), has been on a decline following the implementation of the new GBDT-based model. This study contributes to promoting innovation and competitiveness within financial institutions by encouraging transparency and strengthening the confidence of investors, stakeholders, and regulators such as the Central Bank through the adoption of Artificial Intelligence (AI) tools that detect credit risks early and help prevent systemic crises.