Federated learning with adaptive reliability aware aggregation for privacy preserving healthcare IoT systems
Abstract
Healthcare Internet of Things (IoT) devices generate vast amounts of sensitive patient data, creating significant privacy risks and regulatory challenges when processed through centralized architectures. Federated Learning (FL) offers a promising solution by enabling decentralized model training without sharing raw data. However, real-world deployment faces critical challenges including device heterogeneity, intermittent connectivity, non-independent and identically distributed (non-IID) data, and unreliable node participation. This paper proposes a Reliability-Aware Adaptive Aggregation (RAAA) framework that dynamically evaluates client trustworthiness through three dimensions: update quality (local accuracy and gradient alignment), participation consistency (historical engagement patterns), and communication latency. These metrics are integrated into a composite reliability score using weighted aggregation, which adaptively prioritizes contributions from trustworthy nodes during global model updates. The framework was validated using PhysioNet electrocardiogram (ECG) datasets (MIT-BIH and PTB) with 20 simulated healthcare IoT clients under conditions of node dropout and data heterogeneity. Experimental results demonstrate that RAAA achieved 91.7% classification accuracy on arrhythmia detection, outperforming Federated Averaging (FedAvg) (85.2%), Federated Proximal (FedProx) (86.4%), and Q-FedAvg (87.1%) by 6.5%, 5.3%, and 4.6%, respectively. The proposed method converged 17% faster (43 versus 52 rounds) and maintained an 85.6% client participation rate. Cross-dataset validation on the PTB database confirmed consistent improvements across different physiological signals. The RAAA framework provides a robust, privacy-preserving FL solution for healthcare IoT environments with superior accuracy, faster convergence, and resilience to unreliable nodes. All evaluations are conducted in a simulated environment; while theoretical energy and complexity analyses suggest suitability for resource-constrained edge devices, real-hardware validation remains an important direction for future work.
Related
- Federated learning with update level noise perturbation and blockchain audit logging for privacy-preserving healthcare ECG classification
- TrustLite-FL: multi-statistic trust aggregation for robust federated learning in consumer IoT
- ATLAS: an adaptive threat learning and analysis system using heterogeneous federated deep learning and FedNova aggregation for multi-cloud environments
- Causal-enhanced federated transfer learning framework for privacy-preserving cross-organizational strategic collaboration
- FedEAVE: federated evidence alignment and causal diagnosis for privacy-preserving personalized learning
BibTeX
@article{barot2026federated,
title = {{Federated learning with adaptive reliability aware aggregation for privacy preserving healthcare IoT systems}},
author = {Vishal Barot and Jashvant Dave and Manoj R. and Safvan Vahora and Nilesh Maltare and Jayesh Gangrade},
journal = {Scientific Reports},
year = {2026},
month = oct,
doi = {10.1038/s41598-026-69453-y},
url = {https://www.nature.com/articles/s41598-026-69453-y},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Federated learning with adaptive reliability aware aggregation for privacy preserving healthcare IoT systems
AU - Vishal Barot
AU - Jashvant Dave
AU - Manoj R.
AU - Safvan Vahora
AU - Nilesh Maltare
AU - Jayesh Gangrade
PY - 2026
DA - 2026/10/15
JO - Scientific Reports
DO - 10.1038/s41598-026-69453-y
UR - https://www.nature.com/articles/s41598-026-69453-y
AB - Healthcare Internet of Things (IoT) devices generate vast amounts of sensitive patient data, creating significant privacy risks and regulatory challenges when processed through centralized architectures. Federated Learning (FL) offers a promising solution by enabling decentralized model training without sharing raw data. However, real-world deployment faces critical challenges including device heterogeneity, intermittent connectivity, non-independent and identically distributed (non-IID) data, and unreliable node participation. This paper proposes a Reliability-Aware Adaptive Aggregation (RAAA) framework that dynamically evaluates client trustworthiness through three dimensions: update quality (local accuracy and gradient alignment), participation consistency (historical engagement patterns), and communication latency. These metrics are integrated into a composite reliability score using weighted aggregation, which adaptively prioritizes contributions from trustworthy nodes during global model updates. The framework was validated using PhysioNet electrocardiogram (ECG) datasets (MIT-BIH and PTB) with 20 simulated healthcare IoT clients under conditions of node dropout and data heterogeneity. Experimental results demonstrate that RAAA achieved 91.7% classification accuracy on arrhythmia detection, outperforming Federated Averaging (FedAvg) (85.2%), Federated Proximal (FedProx) (86.4%), and Q-FedAvg (87.1%) by 6.5%, 5.3%, and 4.6%, respectively. The proposed method converged 17% faster (43 versus 52 rounds) and maintained an 85.6% client participation rate. Cross-dataset validation on the PTB database confirmed consistent improvements across different physiological signals. The RAAA framework provides a robust, privacy-preserving FL solution for healthcare IoT environments with superior accuracy, faster convergence, and resilience to unreliable nodes. All evaluations are conducted in a simulated environment; while theoretical energy and complexity analyses suggest suitability for resource-constrained edge devices, real-hardware validation remains an important direction for future work.
SN - 2045-2322
ER -