Penerapan Synthetic Minority Oversampling Technique (SMOTE) terhadap Data Tidak Seimbang pada Pembuatan Model Komposisi Jamu

Abstract

As the times many people use herbal remedies (jamu) to address health issues. Herbal medicines are made from plants with a specific composition to produce certain properties, so a model is needed to be made in order to find the right formula to make herbal medicine with certain properties. In this study, the response being investigated is a potent herbal medicine in treating mood and behavior disorder. In this analysis, the model is developed using logistic regression. The accuracy of the model can be seen from the Area Under Curve (AUC). Imbalanced data on the response variable can cause the value of AUC become low. One of the ways to solve it is using Synthetic Minority Oversampling Technique (SMOTE). From this analysis, Nagelkerke R2 values generated by the model with SMOTE 3.2% lower than model without SMOTE. Nonetheless, the model with SMOTE is more accurate than model without SMOTE because has higher AUC value. The resulting AUC is equal to 0.976 for the model with SMOTE and 0.908 for model without SMOTE. The results show that SMOTE can increase the accuracy of the model for imbalanced data.