기업 신용평가 모형에 설명 가능한 AI 적용을 위한 TabNet 기반 프레임워크: 기업 신용평가 모형 연구를 중심으로

Credit risk management has recently been emphasized through the advancement of the corporate credit evaluation model. Representative models used in studies related to corporate credit evaluation have various statistics and machine learning techniques, and recently, deep learning-based algorithms that show good performance in various fields have been studied to show high predictive performance. However, in corporate credit evaluation research, not only performance but also the content to focus on together is an interpretation of results and the processing of imbalanced data. In this work, we propose a preprocessing method that generates new variables for performance, sampling techniques for handling unbalanced data, and a new framework utilizing TabNet, an explainable artificial intelligence (XAI) based on deep learning. As a result of the study, the high performance of the new framework was confirmed compared to logistic regression and multilayer perceptron. In addition, as a result of analyzing explanatory power, it was confirmed that stability and profitability indicators had a significant effect on prediction, and growth and activity indicators had a negligible effect.