Analisis Regresi Logistik dan CART untuk Credit Scoring dengan Penanganan Kelas Tak Seimbang

Authors

  • Siwi Haryu Pramesti Student
  • Indahwati Indahwati Department of Statistics, IPB University, Indonesia
  • Utami Dyah Syafitri Department of Statistics, IPB University, Indonesia

DOI:

https://doi.org/10.29244/xplore.v11i3.1015

Keywords:

CART, binary logistic regression, SMOTE, credit scoring, credit risk

Abstract

The absence of collateral for a type of credit will increase the bank's credit risk (failed to pay). Banks apply the precautionary principle by managing their credit portfolios so that potential hazards that occur can be measured and controlled in a model. Credit scoring describes how likely a debtor will fail with payments. This study aimed to compare logistic regression analysis and Classification and Regression Trees (CART) in classifying debtors to evaluate credit policies. One of the problems in classification is unbalanced data. Synthetic Minority Oversampling Technique (SMOTE) is a technique to handle the unbalanced problem in classification. The results show that the logistic regression model with SMOTE has higher sensitivity than the CART model, and there was no difference in Area Under Curve (AUC). The variables that have significant effects on the classification of debtors (good, bad) are level of education, homeownership status, and income.

Downloads

Published

2022-09-30

How to Cite

Siwi Haryu Pramesti, Indahwati, I., & Syafitri, U. D. (2022). Analisis Regresi Logistik dan CART untuk Credit Scoring dengan Penanganan Kelas Tak Seimbang. Xplore: Journal of Statistics, 11(3), 226–237. https://doi.org/10.29244/xplore.v11i3.1015