Hybrid VAE-GNN Framework for Real-Time Anomaly Detection in IoT-Based Health Monitoring Systems

A novel Continuous real-time patient monitoring has been made possible by the growth of IoT in the healthcare industry. Yet, physiological variability, data noise, and strict latency constraints make it difficult to accurately spot anomalies within high-dimensional data streams. A new Hybrid Variational Autoencoder-Graph Neural Network (VAE-GNN) Framework for real-time anomaly identification in Internet of Things (IoT)-based health monitoring systems is presented in this study. The VAE component uses a weighted contrastive loss to enforce the separation of normal and anomalous patterns while compressing multivariate health data into a 64-dimensional latent space with a 99.75% reconstruction accuracy. The GNN module increases the sensitivity of anomaly identification by modeling correlations and temporal dependencies using dynamically created patient-specific physiological graphs. The framework's capacity to attain 99.75% accuracy, 99.00% precision, 98.17% recall, and an F1-score of 98.54% is demonstrated through evaluation on a 10,000-sample synthetic health dataset. Robustness is ensured and false positives are reduced via an efficient thresholding technique that is calibrated at the 95th percentile of reconstruction errors. The suggested framework is perfect for practical IoT-based healthcare applications, such as remote monitoring and early chronic illness intervention, since it takes into account scalability, personalization, and low latency needs. This hybrid approach paves the way for scalable, intelligent, and dependable health monitoring systems.