Enhancing IoT Healthcare with Federated Learning and Variational Autoencoder.

Area of Science:

• Healthcare Technology
• Artificial Intelligence
• Data Privacy

Background:

• The Internet of Things (IoT) in healthcare aims to improve patient services using hospital data.
• Privacy concerns hinder data sharing, creating a barrier for effective healthcare solutions.
• Federated learning (FL) enables collaborative model training while preserving data privacy.

Purpose of the Study:

• To address the challenges of data heterogeneity in federated learning for IoT healthcare.
• To propose novel methods for efficient group formation and aggregation weighting in FL.
• To enhance the performance of global model training in decentralized healthcare systems.

Main Methods:

• Utilized an autoencoder for feature extraction and learning latent space divergence to form patient data groups.
• Developed a novel aggregation process incorporating patient data features to optimize global model performance.
• Implemented and compared proposed group formation and aggregation weighting strategies against conventional methods.

Main Results:

• The proposed methods demonstrated superior performance compared to existing conventional approaches.
• Achieved a 20.8% increase in accuracy with the novel group formation and aggregation techniques.
• Showcased a 7% greater reduction in loss compared to traditional federated learning aggregation methods.

Conclusions:

• Novel group formation and aggregation weighting strategies significantly improve federated learning in IoT healthcare.
• The proposed autoencoder-based feature divergence method effectively handles patient data heterogeneity.
• These advancements offer a more efficient and accurate approach to privacy-preserving collaborative model training in healthcare.