STUDI KOMPARASI METODE SVM-SMOTE DAN SMOTE-TOMEK DALAM MENGATASI IMBALANCE CLASS MENGGUNAKAN MODEL XGBOOST PADA KLASIFIKASI RUMAH TANGGA PENERIMA KUR

This study aims to compare the SMOTE, SVM-SMOTE, and SMOTE-Tomek methods using the XGBoost model in overcoming the problem of class imbalance and to determine the factors that affect the status of KUR recipients in West Java Province. Three XGBoost models with class balancing techniques SMOTE, SVM-SMOTE and SMOTE-Tomek were applied to SUSENAS data of West Java Province in 2023 consisting of 1 response variable and 19 predictor variables. The results showed that the XGBoost model with the SMOTE balancing method produced better accuracy in overall data classification, but was less effective in classifying minority classes as reflected by low sensitivity and F1-Score values. The XGBoost model with the SMOTE-Tomek balancing method showed better performance in capturing minority classes with higher sensitivity and F1-Score values. The most influential variables in this model in order are per capita expenditure, urban/rural classification, motorcycle ownership, dwelling wall materials and land ownership. Per capita expenditure has the largest influence on the classification of KUR recipients, indicating that household financial management is a major factor in lending decisions. Urban/rural classification and motorcycle ownership also contributed significantly, reflecting differences in social and economic access between regions. Overall, economic factors, infrastructure and social accessibility are the main considerations in determining KUR recipient households in West Java Province.