Previous research has inadequately explained the relationship between non-financial information and credit risk due to the selective assignment of credit ratings to firms. This study addresses this limitation by employing machine learning techniques trained on firms’ past credit ratings and using the resulting variable as a control to analyze the relationship between non-financial factors and credit risk, measured by the distance to default (DD). The key findings of this study are as follows. First, using a sample of involuntarily delisted firms, we confirm that our trained machine learning model effectively captures default patterns. Second, we find that governance (G) among ESG factors significantly reduces default probability. In contrast, social (S) factor, analyst coverage, and carbon emissions do not show a significant relationship with DD. This study makes several contributions. First, by employing a machine learning framework, our research extends beyond the traditional focus on bond-issuing firms. Second, we highlight that specific ESG factors, particularly governance (G), should be incorporated into credit rating assessments along with traditional financial indicators. Third, our research provides an analytical framework for examining the relationship between involuntarily delisted firms and their financial performance.