← Back to Publications List

An Explainable Communication-Efficient Federated Deep Learning Framework for Privacy-Preserving IoT-Based Smart Healthcare

Students & Supervisors

Student Authors
Marowa Jahan
Bachelor of Science in Computer Science & Engineering, FST
Md. Ridwan Al Mustavy
Bachelor of Science in Computer Science & Engineering, FST
Md. Fahad Hasan Rahman
Bachelor of Science in Computer Science & Engineering, FST
Supervisors
Muhibul Haque Bhuyan
Professor, Faculty, FE

Abstract

The recent development of the Internet of Things (IoT) wearable technologies has allowed the contin-uous monitoring of physiological signals for smart health applications. Traditional deep learning cen-tralization solutions, however, require sending sensitive biomedical data to distant servers, which poses privacy concerns, communication burdens, and limited scalability on diverse edge devices. Current federated learning (FL) approaches address privacy concerns but tend to compromise performance when data is not independent and identically distributed (non-IID), have limited interpretability and incur unnecessary communication overhead during collaborative training. To overcome these issues, in this paper, a novel Communication-Efficient Explainable Federated Deep Learning (CE-EFDL) meth-od is presented for privacy-preserving stress detection using wearable IoT devices. The proposed framework integrates a lightweight one-dimensional convolutional neural network (1D-CNN) on edge devices and an updated FedProx based aggregation strategy to achieve learning stability under diverse client’s distribution while minimizing communication overhead. Moreover, a SHAP-based explainabil-ity module is added to offer clear and clinically comprehensible prediction explanations without re-vealing any patient information. The framework is tested by simulating distributed IoT clients with realistic non-IID data partitions on the publicly available WESAD physiological dataset. Thorough experiments study the classification performance, communication efficiency, convergence behavior, scalability, resource usage and robustness via ablation studies and statistical validation. The experi-mental results confirm the proposed framework's suitability for real-time edge-enabled healthcare ap-plications, providing competitive performance in stress detection and significantly reducing communi-cation cost and preserving data privacy. The proposed framework delivers an integrated approach that simultaneously solves the problems of privacy safeguarding, communication efficiency, heterogeneous federated learning, and explainable AI to offer a practical framework for secure next-generation IoMT systems.

Keywords

Federated Learning Explainable Artificial Intelligence (XAI) Internet of Medical Things (IoMT) Edge Computing Wearable Sensors Privacy-Preserving Machine Learning Communication-Efficient Learning.

Publication Details

  • Type of Publication:
  • Conference Name: International Conference on Innovation and Technopreneurship (ICIT2025)
  • Date of Conference: 09/10/2026 - 09/10/2026
  • Venue: INTI International University, Malaysia
  • Organizer: INTI International University, Malaysia
View on Publisher