설명가능한 인공지능을 활용한 COVID-19 이후 기업신용등급 하락 예측 연구

The COVID-19 pandemic has profoundly impacted the global economy, leaving many firms facing financial difficulties. This study aims to predict corporate credit rating downgrades after COVID-19 by comparing various artificial intelligence (AI) models and interpreting the results using explainable AI techniques, such as Local Interpretable Model-Agnostic Explanations (LIME) and Shapley Additive Explanations (SHAP). Financial data from publicly listed firms in South Korea from 2016 to 2023 were utilized to predict firms at risk of credit downgrades. Key variables included firm size, growth, profitability, activity, stability, cash flow indicators, and market indicators. Model performance was evaluated using accuracy, precision, recall, and F1 scores. Results showed that the CatBoost model outperformed others in the pre-COVID-19 period, while the XGBoost model demonstrated superior predictive performance in the post-COVID-19 period. Using the predictions of these models, global and local analyses through LIME and SHAP visualized the impact of each variable on the predictions and identified the importance of key variables. The findings revealed that, before COVID-19, critical variables included cash flow-to-debt ratio, inventory-to-current assets ratio, interest coverage ratio, and fixed asset ratio, emphasizing cash flow and stability indicators. In contrast, post-COVID-19, significant variables shifted to profitability indicators such as interest coverage ratio, fixed liabilities-to-capital ratio, sales-to-SG&A ratio, and depreciation-to- total cost ratio. This study enhances the explainability of corporate credit rating downgrade prediction models using explainable AI techniques, thereby improving the reliability of predictions. The findings are expected to assist financial institutions and investors in making more informed and trustworthy decisions.