Cyber threats are becoming more complex an‎d diverse, therefore Intrusion Detection Systems (IDS) need to be able to adapt to changing network conditions, concept drift, an‎d data distributions that are very unbalanced. Traditional supervised IDS models function well in static settings, but they donʹt hold up well in dynamic networks an‎d generally have large falsepositive rates. This presentation introduces a hybrid paradigm that combines adaptive clustering an‎d ensemble learning to improve the accuracy, adaptability, an‎d resilience of contemporary Intrusion Detection Systems (IDS). The suggested design uses an adaptive micro-clustering method to find changing traffic patterns an‎d provide cluster-based metafeatures including centroid distance, density, an‎d temporal stability. A cost-sensitive stacked ensemble classifier made up of different base learners then processes these derived features to improve generalization an‎d reduce bias against minority classes. The methodology is assessed utilizing benchmark datasets like CIC-IDS2017, UNSW-NB15, an‎d CICIoT2023, focusing on criteria that extend beyond mere accuracy specifically F1-score, G-Mean, ROC-AUC, an‎d detection latency to guarantee real-time applicability. Recent studies (2022–2025) show that using adaptive unsupervised representation with supervised ensembles together leads to big improvements in drift tolerance, recognizing zero-day attacks, an‎d computational efficiency. This study thus aids in the creation of a scalable an‎d intelligent IDS framework capable of selfadaptation an‎d consistent performance in varied, non-stationary network environments

10/28/2025 12:00:00 AM

88022

1404/08/10

Adaptive Clustering-Based Intrusion Detection System for Network Security Using Ensemble Learning

(سيستم تشخيص نفوذ تطبيقي ​​(IDS)، يادگيري جمعي، خوشه‌بندي تطبيقي، تغيير مفهوم؛ امنيت شبكه، عدم تعادل كلاس.)

(Adaptive Intrusion Detection System (IDS), Ensemble Learning, Adaptive Clustering, Concept Drift; Network Security, Class Imbalance.)