The COVID-19 pandemic affects whole segments of the world, including the banking and credit sectors. The banking and credit sector plays an essential role in the Indonesian economy structure due to the function as a collector and channel of funds by creating products offered to people who need to use the credit services. We compare the accuracy of two different machine learning methods - Light Gradient Boosting Machine (LightGBM) and Classification and Regression Tree (CART). We apply this method to classify data and determine factors that affect credit payments in debtor data at PT BPR Syariah Gebu Prima Medan, which contains current credit and bad credit debtors. Based on the evaluation conducted on the classification of factors that affect credit payments, findings showed that the factors that influence the CART method are customers who have a maximum total income of IDR 27,750,000 with a ceiling of more than IDR 57,500,000 are customers who have maximum family members. 3.5 and the LightGBM method is the ceiling, total income, family members, age, and gender with importance values of 65200, 65100,13000,9800, and 4200. However, the CART method has a higher accuracy rate of 85.9% than the LightGBM method is 81%.

Keywords: Machine Learning, Light Gradient Boosting Machine (LightGBM), Classification and Regression Tree (CART).