Next-Generation Digital Twin Model with Unobtrusive RF Multi-Sensing for AI-Based Human Monitoring

In healthcare, Digital Twin (DT) model can create real-time digital representations of patients’ physiological states, enabling intelligent, data-driven decision-making for continuous monitoring and personalized treatment-without the invasiveness of traditional methods. This study presents a next-generation RF-powered DT model, integrating AI-driven analytics with multi-vital human sensing to establish a virtualized, self-learning, and adaptive framework for health monitoring. Leveraging unobtrusive Radio Frequency (RF) sensors, advanced signal processing, and Artificial Intelligence (AI), the proposed DT model facilitates continuous, non-contact assessment of respiration and exhaled hydration, ensuring a scalable, cost-effective, and efficient
approach to real-time physiological monitoring. This work integrates unobtrusive ESP32 Wi-Fi sensors with Channel State Information (CSI) for respiration tracking and flexible UWB RF sensors for hydration monitoring. Enhanced signal processing achieves 100% accurate estimation of Breaths Per Minute (BPM) within the accepted inter-observer variability threshold (±5
BPM) from raw respiration data. To overcome the challenge of limited real-world respiration and hydration datasets, novel statistical augmentation methods are employed to generate synthetic data, validated through cross-correlation techniques (Pearson, Kendall, and Spearman). AI models, including supervised and semi-supervised Machine Learning (ML) and Deep Learning (DL), are implemented for binary and multi-class classification of respiration and hydration data. Random Forest achieves top accuracies of 88% (binary) and 69% (multi-class) in semi-supervised classification, while Decision Tree attains
89% and 83% accuracy in supervised exhaled hydration classification. Additionally, K-Nearest Neighbors (KNN) achieves 93% (binary) and 89% (multi-class) accuracy in semi-supervised hydration classification.