Banking Credit Risk Analysis with Naive Bayes Approach and Cox Proportional Hazard

Abstract

Credit is needed for some people for certain purposes. In credit, it takes a party that can be used as an intermediary such as a bank. The debtor may not be able to make payments according to the original policy or even cause losses where the Bank may lose the opportunity to earn interest, causing a decrease in total income. This problem is included in the case of non-performing loans. In statistics, the duration of time between a person not making a payment on time until a non-current loan occurs can be predicted using survival analysis. Meanwhile, to predict credit status, you can use classification or prediction methods in machine learning to find out how much influence the predictor variable has. In this study, with a different case, focusing on the credit risk case of how a bank decides to provide credit to prospective debtors using the classifier method found in Machine Learning, namely Naive Bayes and Cox regression from survival analysis. Through the evaluation test of the naive bayes classifier model using accuracy values, confusion matrix and ROC, it can be concluded that this model is a model with good performance for predicting credit status. Multinomial nave Bayes in this study has a higher performance value than Gaussian Naïve Bayes and Bernoulli Naïve Bayes which is 92%. Through cox regression, it is obtained that income factors and loan history have a major influence on determining credit status.