A Lightweight, AI-Enhanced Intrusion Detection System for Wireless Body Area Networks Using Unsupervised Anomaly Detection

Abstract

The proliferation of Wireless Body Area Networks (WBANs) in remote healthcare monitoring has introduced unprecedented capabilities for patient care but also significant security vulnerabilities with life-threatening implications. Traditional intrusion detection systems (IDS), often relying on static signatures or simplistic rule-based anomaly detection, are ill-equipped to handle the dynamic and sophisticated nature of modern cyber threats, particularly zero-day attacks. This paper introduces a novel, hybrid, AI-enhanced IDS framework designed for the resource-constrained WBAN environment. The proposed system integrates a lightweight, on-node filtering mechanism with a powerful, centralized, unsupervised convolutional autoencoder (CAE) deployed on the network coordinator. This architecture leverages a rich feature set, including physiological time-series data and a sophisticated energy consumption profile, to achieve robust detection. Unlike conventional models that depend on arbitrary thresholds, our approach employs a data-driven anomaly detection mechanism based on the CAE’s reconstruction error, enabling it to identify both known attack patterns and novel anomalies with high fidelity. The framework is validated using a benchmark physiological dataset and simulated attacks. The results demonstrate a significant advancement over baseline models, achieving a superior F1-Score of 0.96 and an Area Under the Curve (AUC) of 0.98, showcasing its efficacy and potential for securing next-generation medical WBANs.