FedHealthFog: A federated learning-enabled approach towards healthcare analytics over fog computing platform

Area of Science:

• Healthcare technology
• Distributed computing
• Internet of Things (IoT)

Background:

• Federated learning (FL) in healthcare enhances patient data privacy on diverse computing platforms.
• Resource-constrained IoT healthcare systems face challenges with traditional FL due to high compute and communication costs.
• Existing FL models struggle with inefficient training due to centralized data aggregation.

Purpose of the Study:

• To introduce the FedHealthFog framework for efficient, privacy-preserving federated learning in resource-limited healthcare IoT.
• To address the limitations of conventional FL, specifically high latency and energy consumption.
• To optimize distributed learning in delay-sensitive healthcare applications.

Main Methods:

• Developed the FedHealthFog framework, utilizing fog nodes as local aggregators in the FL architecture.
• Implemented a greedy heuristic technique to select optimal fog nodes for global aggregation.
• Evaluated FedHealthFog against benchmark algorithms on key performance metrics.

Main Results:

• FedHealthFog achieved significant reductions in communication latency (up to 87.01%) and energy consumption (up to 57.98%).
• The framework demonstrated improved efficiency compared to state-of-the-art alternatives.
• Reduced the number of global aggregation cycles required for model training.

Conclusions:

• FedHealthFog effectively overcomes challenges in resource-constrained healthcare IoT environments.
• The framework offers a promising solution for privacy-preserving and efficient federated learning.
• FedHealthFog has the potential to revolutionize delay-sensitive applications within IoT ecosystems.