Penerapan Algoritma Extreme Gradient Boosting (Xgboost) Untuk Analisis Risiko Kredit

Based on data on outstanding online loans, in March this year non-current credit (30-90 days) jumped by IDR 3.58 trillion (71.13% YoY) with the highest individual loans amounting to IDR 3.3 trillion and the rest from business entities. Then those experiencing bad credit exceeding 90 days in March 2023 amounted to IDR 1.43 trillion of the total remaining loans outstanding by debtors. This value increased significantly by IDR 1.35 trillion (65.33%) compared to the previous period (YoY), where IDR 1.14 trillion was bad credit from individuals. This happens because it starts from a lack of accuracy in the customer screening process. This research aims to classify potential debtors using data mining techniques using the XGBoost algorithm. The method or steps to achieve the objectives of this research is the Knowledge Discovery in Database (KDD) methodology, which consists of five steps, namely data selection, data preprocessing, data transformation, data mining, and evaluation. The dataset consists of 1000 rows of data consisting of 700 good and 300 bad, with 21 variables (V1 to V20) while V21 is the target or output. To analyze the data, 10 model scenarios were built to identify the best model. The model results show that the model performance is getting better after SMOTE is carried out with the accuracy and AUC values increasing. The best model was obtained in scenario 1 (90% train data and 10 test data) with an accuracy value of 0.83 and AUC 0.918. The model evaluation results show that the XGBoost algorithm can be used to analyze credit data before accepting/rejecting a credit application.