Credit Risk Prediction System For MSME Loan Process

BPR XYZ is one of the BPR (Rural Bank) located in West Java - Indonesia, that has two primary businesses, which are: (1) collecting funds from customers in the forms of savings and deposits and (2) providing loans to customers. BPR XYZ successfully disbursed approximately more than IDR 300 billion micro-loans to the Micro, Small and Medium Enterprise (MSME) segment and will keep increasing gradually in 2022. Their Non-Performing Loan (NPL) rate in 2021 was above 5% (five percent), more significant than the national standard, and most likely to increase in 2022. This happens because many of their MSME customers cannot pay the debt. This study will conduct predictive analysis using Naïve Bayes and K-Nearest Neighbors algorithm (K-NN) to predict the NPL. From our experiments using historical data and five classes of NPL, we found that naïve bayes do not perform well, with average accuracy just only 19%; meanwhile, K-NN obtains a performance of 59%. Random Forest gained 52%, and SVM with 48%. We tried to restructure the NPL classification label to only two classes with these results. We got better results where K-NN algorithm gained the best performance with an accuracy level of 74%.