A Comparative Study of ADASYN and SMOTE-ENN for Class Imbalance in Intrusion Detection Systems
Pertanika Journal of Artificial Intelligence · 30.10.2026 · 1 (1)
DOI: 10.47836/pjai.1.1.06 ↗Kurz gefasst
Fachartikel in der Zeitschrift Pertanika Journal of Artificial Intelligence, online erschienen am 30.10.2026. Verfasst von Kurniawan Rama Wahyu und 2 weiteren Autor:innen.
Abstract (Originalsprache)
Class imbalance remains a major challenge in multiclass Network Intrusion Detection Systems (NIDS), particularly when rare attack categories are insufficiently represented during model training. Although oversampling techniques are widely used to address this issue, their effectiveness may vary depending on the sampling strategy, classifier, and distribution of attack categories. This study compares Adaptive Synthetic Sampling (ADASYN) and Synthetic Minority Over-sampling Technique combined with Edited Nearest Neighbors (SMOTE-ENN) to investigate their effects on multiclass intrusion detection using the UNSW-NB15 dataset. ADASYN was selected for its adaptive generation of synthetic samples in difficult-to-learn minority-class regions, whereas SMOTE-ENN combines minority-class oversampling with neighborhood based data cleaning to reduce potentially noisy or overlapping instances. Their effectiveness was evaluated against an unbalanced baseline using Random Forest (RF), Decision Tree (DT), Extreme Gradient Boosting (XGBoost), and a hybrid Convolutional Neural Network–Long Short-Term Memory (CNN-LSTM) model. The experiments assessed accuracy, precision, recall, and F1-score using the held-out test set. The results indicate that the impact of resampling is classifier-dependent. CNN-LSTM combined with SMOTE-ENN achieved the highest reported overall performance, with 90% accuracy, 87% precision, 83% recall, and 84% F1-score. These findings demonstrate the potential benefits of combining hybrid resampling with deep learning for multiclass NIDS, while highlighting the importance of evaluating sampling strategies across different classification models and attack categories.
Zitieren
APA
Wahyu, K. R., Husin, N. A., & Fadil, P. F. N. I. M. (2026). A Comparative Study of ADASYN and SMOTE-ENN for Class Imbalance in Intrusion Detection Systems. Pertanika Journal of Artificial Intelligence, 1(1). https://doi.org/10.47836/pjai.1.1.06
Vancouver
Wahyu KR, Husin NA, Fadil PFNIM. A Comparative Study of ADASYN and SMOTE-ENN for Class Imbalance in Intrusion Detection Systems. Pertanika Journal of Artificial Intelligence. 2026;1(1). doi:10.47836/pjai.1.1.06
BibTeX
@article{Wahyu2026,
author = {Wahyu, Kurniawan Rama and Husin, Nor Azura and Fadil, Puteri Fatin Nurul Izzati Mohamed},
title = {A Comparative Study of ADASYN and SMOTE-ENN for Class Imbalance in Intrusion Detection Systems},
journal = {Pertanika Journal of Artificial Intelligence},
year = {2026},
volume = {1},
number = {1},
doi = {10.47836/pjai.1.1.06}
}
EUVIMED