Adaptive resource aware and privacy preserving federated edge learning framework for real time internet of medical things applications.

The Internet of Medical Things requires frameworks that ensure secure processing, computational efficiency, and scalability for continuous healthcare data streams. Existing solutions remain limited in their ability to support real-time anomaly detection, resource-constrained optimization, and multi-tier aggregation in large-scale deployments. This work proposes an Adaptive Federated Edge Learning framework specifically designed for healthcare systems enabled by the Internet of Medical Things. The framework incorporates Adaptive Modular Learning Units that allocate computations according to device-specific resource budgets and update models through federated gradient descent. Dynamic Data Encoding transforms heterogeneous medical data into a unified and decorrelated feature representation with alignment and quality weighting to preserve consistency across distributed sources. A Hierarchical Federated Aggregation mechanism performs device, regional, and global parameter updates using data-size weighting and delay-aware factors to maintain training stability and scalability in heterogeneous network environments. Privacy-Preserving Secure Enclaves provide encrypted model training and aggregation with differential privacy noise injection, ensuring that sensitive clinical data remain protected throughout the learning cycle. Real-time anomaly detection is achieved through a streaming pipeline based on sliding windows, dimensionality reduction, covariance analysis, and adaptive thresholding with context-aware clustering. Experimental evaluation using clinical datasets and simulated Internet of Medical Things environments demonstrated that the framework achieved 96.3% accuracy in controlled conditions and sustained 110 ms latency in streaming anomaly detection.