Continuous and reliable monitoring of soldiers’ health in remote and resource-constrained environments is essential for operational readiness and timely emergency interventions. Long Range Wide Area Network (LoRaWAN) provides an energy-efficient, long-range communication solution for wearable Internet of Medical Things (IoMT) devices, but dynamic factors such as packet loss, latency, and bandwidth limitations can compromise real-time health inference. This paper presents HEAL, Health-Enhanced Adaptive LoRaWAN, an AI-enabled IoMT framework designed to predict soldiers’ health under dynamic physiological and communication conditions. “Adaptive” denotes AI-level robustness, where models sustain stable and accurate health predictions under varying communication constraints and transmission parameters. HEAL employs an end-to-end workflow in which a publicly available multivariate physiological time-series dataset is preprocessed, structured, and automatically labeled using transformer-based large language models. Multiple Deep Learning (DL) architectures are trained and compared to identify models suitable for real-time health status classification, with the best-performing model fine-tuned and deployed in a LoRaWAN simulation to evaluate AI inference under realistic communication constraints. Experimental results show that a BiLSTM model achieves up to 94% accuracy and 95% Macro F1-score on structured physiological data. Within the LoRaWAN simulation, the fine-tuned model maintains approximately 93% accuracy, while the network achieves a packet delivery ratio of up to 91% under favorable conditions (SF7–SF8, 200–400 m). HEAL provides a scalable and energy-efficient framework for systematically evaluating AI models alongside communication reliability metrics, demonstrating that accurate AI-driven health monitoring can be sustained despite dynamic LoRaWAN effects and bridging the gap between model development and realistic operational deployment.
Alghamdi et al. (Thu,) studied this question.