데이터마이닝 기법을 이용한 중년층의 가계부채 연체 가능성 분류 연구

초록

This study aims to detect households at high risk of insolvency among middle-aged debtors using The Survey of Household Finances and Living Conditions in 2019. This study assesses the performance of classifier such as the Decision Tree model in machine learning. It is well known that when the proportion of one class in a dataset is dominant, the prediction performance of classifiers becomes problematic. In order to address the degree of imbalance of two classes in data sets, the ROSE (random oversampling examples) technique was considered. It was found that the ROSE improved the sensitivity and AUC, helping to improve classification prediction accuracy while avoiding overfitting problems. In addition, 1) those with debt-to-asset ratio greater than 0.8, 2) those with debt-to-asset ratio greater than 0.4 and less than 0.8 and got a loan through a savings bank, 3) those with debt-to-asset ratio less than 0.4, non-regular workers, those who did not own a house and those who got a loan for a business were more likely to be delinquent on their debt payment. This study found that the level of Debt to Asset ratio, Debt to Financial Asset ratio, the kind of financial institution, and reason for borrowing money were significant factors of the payment delinquency.

키워드