인공지능을 활용한 금융취약층 신용위험 분류 및 특성 분석 = Classifying and Analyzing Individual Credit Risk among the Financially Vulnerable Using Artificial Intelligence

다국어 초록 (Multilingual Abstract)

This study aims to develop an artificial intelligence model that predicts individual credit risk. The data for this study was administrative data of new applicants for debt relief programs between January 2014 and May 2024. The sample includes a total of 791,393 individuals. Three supervised learning algorithms—neural network model, random forest model, and naive Bayes classification model—were employed to evaluate model performance. The naive Bayes classification model was selected as the optimal model. The main findings are as follows: First, the accuracy of identifying credit risk prediction using machine learning was 86.19%. Accuracy here refers to the proportion of correctly predicted cases in the overall dataset, including both the sensitivity(how well the model identifies cases that are actually poor credit) and the specificity(how well the model identifies cases that are actually good credit). Second, the key characteristics associated with poor credit include being in the 20s or 30s age group, having a college degree or higher, being self-employed or in temporary employment as compared to full-time employment, being unmarried, having more dependents, and having many debts. This study demonstrates that credit risk—a key factor in financial activities and economic recovery—can be effectively and efficiently identified through artificial intelligence-based classification and analysis. Furthermore, this study enhances our understanding of the factors related to credit risk among users of debt relief programs and helps identify at-risk groups that may need support to prevent deteriorating credit risk.