Trust-Aware Federated Meta-Learning with Adaptive Privacy and Explainable AI for IoT Healthcare
The increasing deployment of connected sensing devices in healthcare enables distributed physiological monitoring and data-driven clinical support, but centralized collection of sensitive records creates substantial privacy and security concerns. This paper presents a Trust-Aware Federated Meta-Learning with Adaptive Privacy and Explainable AI (TAFML-XAI) framework for heterogeneous IoT healthcare environments. The framework combines trust-weighted aggregation, adaptive Gaussian differential privacy, Model-Agnostic Meta-Learning (MAML), and federated SHAP-based explanations. The trust scores are computed based on the divergence between each client update and the global model, while privacy-preserving perturbations (modes of differential privacy) are imposed on the updates before sending. The revised formulation explicitly analyzes the interaction between privacy noise, trust estimation, and explanation stability. Experiments reported in the manuscript use the MIT-BIH Arrhythmia, PPG-DaLiA, and UCI Air Quality datasets. The reported classification accuracy of the proposed framework is 98.61%, while robustness experiments report 97% and 94% accuracy with 10% and 20% noisy/adversarial clients, respectively. The reported meta-learning experiment reduces convergence from 120 to 75 communication rounds and increases personalization accuracy from 88% to 97%. SHAP evaluation reports fidelity of 96.2%, consistency of 94.8%, and feature-importance correlation of 93.5%. The results show that the integrated framework can improve robustness, personalization, privacy-utility control, and interpretability; however, the centralized baseline and resource-overhead measurements require controlled reruns before the final submission.
Authors
- Sirisha Rani Kolli
- Md Zia Ur Rahman
Institutions
- Koneru Lakshmaiah Education Foundation (IN)
Publication Details
- Journal
- International Journal of Electronics and Communication Engineering
- Published
- 2026-11-12
- DOI
- https://doi.org/10.14445/23488549/ijece-v13i10p111
- Primary Topic
- Privacy-Preserving Technologies in Data
- Type
- article
- Field-Weighted Citation Impact
- 0.00