SMSE loans have an important role on improving the regional economy. The problems of SMSE loan services faced by banks today are (1) brief processing time, and (2) maintaining loan quality and increasing SMSE loan portfolios. To solve those problems, a regulatory comply model for predicting default SMSE loan application is needed. This paper presents a machine learning model that complies Indonesian regulatory NPL limit on Bank XYZ case study. The modeling uses 38,066 datasets obtained from Bank XYZ loan data. Data is processed on XGBoost algorithms, Random Forest and Artificial Neural Network (ANN). However, this study explores further on algorithms capability to obtain accuracy and reduction of NPL (Non-Performing Loan) Rate Performance using Machine Learning (ML). XGBoost has become best model in this experiment with 95.54% accuracy and 3.63% Model’s NPL, which is below the existing Bank XYZ’s business model NPL 5.25% and Indonesian NPL regulation 5%.