Adaptive federated ensemble learning with hyperparameter optimization for distributed healthcare IoT

Abstract

The rapid evolution of IoT-enabled healthcare systems has resulted in large-scale distributed data generation across various institutions. While federated learning enables collaborative model training without centralized data, it degrades non-identical data distributions and statistical imbalance across clients. Moreover, the convergence instability and hyperparameter sensitivity remain critical challenges in heterogeneous healthcare environments. To address these limitations, this paper proposes a Stability-Aware Secure Federated Ensemble Framework integrated with LampyraCetacea Optimization (ASEFE-LCO). The platform integrates FedAvg and FedProx algorithms with an adaptive multi-model ensemble mechanism to ensure convergence stability under non-IID conditions. To enhance predictive robustness, LampyraCetacea Optimization is employed for dynamic hyperparameter adaptation in non-convex environments. To offer formal privacy tamper-proof protection, an integrated blockchain layer is applied to ensure decentralized auditability. The proposed architecture offers an end-to-end intelligent platform that combines optimization stability, security assurance, scalability, and distributed knowledge integration. As a core tenet of the proposed model, the experimental evaluation of the IoT healthcare security benchmark dataset achieves 99.94% accuracy, indicating a robust, secure, and clinically interpretable solution for an intelligent healthcare system.