Intrusion Detection System for Secure IoT-Based Healthcare Systems Based on Temporal Graph Attention

Abstract

The rapid integration of Internet of Things (IoT) technologies into healthcare infrastructures has significantly enhanced real-time monitoring and clinical automation, but it has also introduced complex cybersecurity vulnerabilities. Traditional intrusion detection systems (IDS) typically network flows independently and fail to capture relational dependencies among communicating medical devices. This limitation reduces their effectiveness against coordinated and multi-stage cyberattacks in IoT healthcare environments.

This paper proposes a temporal graph attention-based intrusion detection framework that models IoT healthcare traffic as dynamic communication graphs constructed using sliding time windows. In the proposed approach, nodes represent IoT devices, while edges represent communication flows enriched with statistical network features. A Graph Attention Network (GAT) is employed to learn structural dependencies and generate graph-level embeddings for window-based attack detection. To ensure realistic evaluation and avoid temporal leakage, a strict time-based data splitting strategy is adopted.

Experimental results demonstrate that the proposed graph-based framework consistently outperforms conventional machine learning baselines, including Random Forest, XG-Boost, and Multi-Layer Perceptron, in window-level intrusion detection tasks. Ablation analysis further confirms the impact of temporal window size on detection performance. The proposed method provides a scalable and intelligent security solution suitable for real-time IoT-enabled healthcare systems.