SYSTEMATIC LITERATURE REVIEW: PENGARUH KARAKTERISTIK DATASET TERHADAP EFEKTIVITAS VARIAN SYNTHETIC MINORITY OVERSAMPLING TECHNIQUE: SYSTEMATIC LITERATURE REVIEW: THE EFFECT OF DATASET CHARACTERISTICS ON THE EFFECTIVENESS OF SYNTHETIC MINORITY OVERSAMPLING TECHNIQUE VARIANTS

Dwiki Likuisa (Universitas Amikom Purwokerto)
Imam Tahyudin (Amikom Purwokerto University)
Berlilana (Amikom Purwokerto University)

RABIT: Jurnal Teknologi dan Sistem Informasi Univrab
Vol 11 No 2 (2026): Juli

Class imbalance is a major challenge in machine learning, particularly in datasets with varying overlap and dimensionality. While SMOTE and its variants offer synthetic oversampling solutions, their comparative effectiveness across diverse data structures requires synthesis. This systematic literature review applies the PRISMA 2020 protocol using a single database source, ScienceDirect, analyzing 20 papers (2021 to 2025) that explicitly compare at least two SMOTE variants (SMOTE, Borderline-SMOTE, ADASYN, SMOTE-ENN) on public or well-documented benchmark datasets. Of 777 identified articles, a staged screening process yielded 20 papers meeting the inclusion criteria. Results show that SMOTE-ENN and noise-cleaning based approaches consistently outperform in multiclass datasets with high overlap, while ADASYN and adaptive-weighting approaches excel in datasets with extreme imbalance ratios but low overlap. Borderline-SMOTE and boundary-based variants provide stability under moderate imbalance with low-to-medium dimensionality. Furthermore, G-Mean, F1-Score, and AUC proved more informative than accuracy for imbalanced conditions. This review offers methodological recommendations for selecting oversampling strategies based on dataset characteristics.