Analisis Dampak SMOTE terhadap Feature Importance pada Klasifikasi Data Migraine menggunakan Random Forest dan Extra Trees

Henny Leidiyana (Universitas Bina Sarana Informatika)

Jurnal Komtika (Komputasi dan Informatika)
Vol. 10 No. 1 (2026)

This study analyzes the impact of the Synthetic Minority Over-sampling Technique (SMOTE) on model performance and feature importance in the classification of migraine patients using Random Forest (RF) and Extra Trees (ET) algorithms. Evaluation was conducted based on recall and F1-Score for the minority class, as well as Permutation Importance analysis. The results indicate that ET, especially when combined with SMOTE (ET + SMOTE), delivers the best performance for the minority class. ET + SMOTE achieved an average F1-Score of 0.7000 and an average recall of 0.8041 using 11 optimal features, indicating better feature efficiency. The application of SMOTE significantly affected the ranking of important features. Although SMOTE improved detection for some minority classes, its impact was not always consistent and occasionally reduced performance on other minority classes. This study concludes that SMOTE alters feature contributions and model interpretability, as well as enhances performance on certain minority classes, particularly when combined with ET.