Credit Scoring Refinement Using Optimized Logistic Regression

A poor credit scoring model will give a poor power for predicting defaulted loan. There are many approaches for modeling the default prediction, such as classical logistic regression and Bayesian logistics regression. In this paper, we applied both classical logistic regression and AUC (Area under Curved) optimized using Nelder-Mead Algorithm for refining a credit scoring model that has already been used for several years by an International bank in Indonesia. Both classical logistics regression and AUC optimized method perform well in improving the model, but logistic regression still better in some aspects. AUC Optimized model has higher AUC than logistic regression model but has lower Kolmogorov-Smirnov Score.