ADML-IDS: An Adaptive Ensemble Machine Learning
Framework for Intrusion Detection in Wireless Ad Hoc and
Sensor Networks
Ahmed Aziz1,∗, Mahmoud Abdel-Salam2
1Dean of the Engineering School, Central Asian University, Uzbekistan
2Faculty of Computer and Information Sciences, Mansoura University, Egypt
Emails: a.ahmed@centralasian.uz; masalam99@yahoo.com
Abstract
As wireless sensor networks (WSNs) and mobile ad hoc networks (MANETs) are (DoS) attacks
has become a critical security concern in mission-critical wireless (DoS) attacks has become
a critical security issue. This paper proposes ADML-IDS, an Adaptive Machine Learning
Intrusion Detection System that integrates ensemble of Random Forest, XGBoost and Gradient
Boosting classifiers using a Flooding, and Scheduling—as well as normal traffic. Flooding,
Scheduling and normal traffic. Experiments are conducted on the open-source WSN-DS dataset,
which contains 166,000 network observations using the LEACH hierarchical routing protocol
with 23 features obtained from NS-2 simulation. The data preprocessing steps include Min-
Max normalisation and Synthetic Minority Over-Sampling Technique (SMOTE) to balance
classes, and importance-based feature selection to retain 19 features. A rigorous ten-fold crossvalidation
strategy is followed. ADML-IDS achieves an overall accuracy of 99.57%, weighted
F1-score of 0.9956 and AUC-ROC of 0.9985. AUC-ROC of 0.9985, outperforming each of the
sub-classifiers and five state-of-the-art methods. Scalability experiments demonstrate that the
accuracy of detection remains above network size reaches 200 nodes, and with a reasonable
computational cost. A formal presentation of the energy-aware network model and ensemble
decision rule is tables are also included along with a full description of the algorithm. tables.
Keywords: Wireless sensor networks; Ad hoc networks; Intrusion detection system; Ensemble
machine learning; DoS attacks; LEACH protocol; WSN-DS dataset; Random Forest; XGBoost;
Soft-voting classifier