DefinePK

Abstract

Wireless Sensor Networks (WSNs) have become vital for diverse applications such as military monitoring, healthcare, and urban traffic analysis. However, challenges like limited battery power, overlapping coverage, and energy dissipation hinder their performance and security. Traditional intrusion detection methods, including rule-based and cryptographic approaches, often struggle with adaptability or computational overhead in resource-constrained WSNs. Deep learning models, while effective, are typically too heavy for real-time deployment. To overcome these issues, this study proposes a stacked ensemble machine learning framework combining Decision Trees, Random Forest, XGBoost, and SVM classifiers. This approach leverages the strengths of multiple models via a meta-classifier to improve threat prediction accuracy, adaptability, and energy efficiency. Evaluated on standard WSN intrusion detection datasets, the framework achieves over 99.7% accuracy with high F1-scores and ROC-AUC, demonstrating superior detection of attacks like Blackhole, Flooding, Grayhole, and TDMA. The results highlight the method’s potential for scalable, lightweight, and robust real-time WSN security applications