A Lightweight Temporal Gated Recurrent Unit Using Priority-driven Attention Mechanism for Secured Internet of Healthcare Things Paradigm

— Healthcare 4.0 is the keystone of the Fourth Industrial Revolution, which connects people and medical devices with real-time data analytics and communication systems. The advent of the Internet of Healthcare Things (IoHT) has enabled real-time patient monitoring, better decision-making, and improved quality of service. Conversely, the growing connectivity of healthcare equipment also divulges serious susceptibilities, making the IoMT paradigm prone to different cyberattacks. In this study, we propose a Lightweight Temporal Gated Recurrent Unit (LTGRU) for anomaly detection in resource-constrained edge and gateway devices of healthcare networks. The LTGRU-based Intrusion Detection System (IDS) contains (i) a lightweight Gated Recurrent Unit (GRU) by decreasing the hidden module size to reduce the number of trainable parameters. This module extracts time-based traffic patterns and differentiates normal and attack network activities with improved accuracy, and (ii) an adaptive ReLU activation function combined with a Priority-driven Temporal Attention Mechanism (PTAM) to replace the tanh activation function of the original GRU. This module is used to improve efficiency and focus on critical time steps in IoHT data streams. The proposed model efficiently captures temporal patterns without leading to significant computational overhead. Besides, this model minimizes storage and processing costs compared to traditional deep networks. Compared to conventional IDS models, our LTGRU considerably decreases parameter count from 61824 to 6240, achieving nearly a 90% reduction in model size compared to conventional GRU, making it the perfect solution in low‑power IoHT edge devices. The proposed LTGRU-IDS is efficiently implemented on the Edge‑IIoTset dataset using Google Colab. The experimental fallouts prove that the proposed IDS outstrips other models with better performance of 98.66% accuracy, 98.19% sensitivity, 98.96% specificity, 98.78% precision, 1.18% false negative rate, and 1.84% false positive rate, respectively. Statistical validation using the Wilcoxon signed-rank test proved that these performance measures were statistically significant related to other prevailing IDS models.
Keywords— cyberattacks; Gated recurrent unit; healthcare IoT; Industrial Internet; Priority-driven attention;