Comparison of Logistic Regression and XGBoost for Predicting Potential Debtors

Abstract

Income earned from loan interest is the main income for banks, especially conventional banks. However, currently the population ratio between total customers and customers who have credit facilities is very far away. Moreover, the marketing strategy undertaken still relies on the referral system, so it does not reach potential debtors who have no relations with other debtors. This study aims to predict the potential of customers in taking credit facilities or becoming a debtor. The study is conducted by comparing logistic regression model and XGBoost. The accuracy of the testing data on the logistic regression model is 88% while the XGBoost is 92%. The comparison results show that the XGBoost method has better results based on four evaluation indicators namely accuracy, sensitivity, specificity, and precision. Because those values on the XGBoost are greater than logistic regression.