Federated Learning and Blockchain-Enabled Fog-IoT Platform for Wearables in Predictive Healthcare

Marc Jayson Baucas, Petros Spachos, Konstantinos N. Plataniotis

TL;DR

This paper introduces a novel fog-IoT platform that combines federated learning and private blockchain to enhance data privacy, service integrity, and network adaptability for wearable IoT devices in predictive healthcare. By decentralizing learning to the fog layer and securing data access with blockchain, the platform aims to overcome the limitations of current e-health systems.

What Problem Does This Paper Address?

The increasing adoption of wearable IoT devices in healthcare offers immense potential for predictive analysis and early diagnosis. However, several challenges hinder their widespread and secure implementation. Firstly, the sensitive nature of the collected physiological data raises significant privacy concerns. Secondly, the limitations of wearable devices themselves can lead to vulnerabilities and potential data leakages. Thirdly, ensuring the integrity and accuracy of the data and the predictive models is crucial for reliable healthcare services. Finally, the heterogeneity of wearable IoT devices and the rapid pace of technological advancements create challenges in network adaptability and scalability. This paper aims to address these interconnected issues by proposing a robust and secure platform.

Methods

The core of the proposed solution lies in the integration of three key technologies within a fog-IoT architecture:

  • Federated Learning (FL): Instead of sending raw data to a central server, FL allows local models to be trained on wearable devices (or at the fog layer). Only the model updates (weights) are then aggregated centrally. This significantly enhances data privacy by keeping raw patient data localized. The paper leverages FL to aggregate local models from fog nodes into a global model, ensuring continuous learning and adaptation.
  • Private Blockchain: A private blockchain is employed to manage client authorization and store training information. This creates a secure, immutable ledger for tracking model updates and ensuring that only trusted devices can access the predictive model. The use of a private blockchain is chosen over public blockchains due to the lower processing requirements, making it suitable for resource-constrained IoT environments.
  • Fog-IoT Architecture: The platform utilizes a fog-IoT architecture, which brings computation and data storage closer to the edge devices. This is particularly beneficial for offloading the computationally intensive task of model training from low-power wearable devices to more capable fog nodes. This distributed approach enhances network adaptability and allows for efficient resource reallocation.

Datasets

The paper uses the Human Activity Recognition (HAR) Using Smartphones Data Set from the UCI Machine Learning Repository. This dataset, collected from 30 volunteers performing six activities (walking, walking upstairs, walking downstairs, sitting, standing, and lying down) using a Samsung Galaxy SII smartphone, provides 3-axial linear acceleration and 3-axial angular velocity data. The authors chose this dataset and a 1D Convolutional Neural Network (CNN) for Human Activity Recognition to simulate patient data for training HAR models, focusing on the federated learning aspect rather than maximizing classifier accuracy.

Key Insights and Ideas

  • Decentralized Learning for Privacy: The fundamental insight is that federated learning, by keeping raw data local, is a powerful tool for preserving patient privacy in wearable IoT healthcare.
  • Blockchain for Trust and Access Control: Integrating a private blockchain provides a secure and auditable mechanism for managing device access and tracking model evolution, crucial for maintaining trust in a healthcare context.
  • Fog Computing as an Enabler: The fog layer acts as a critical intermediary, enabling efficient offloading of computationally intensive tasks like model training from resource-constrained wearables, thereby improving adaptability and scalability.
  • Optimized Aggregation for Performance: The paper proposes a scaling scheme for federated learning where local models with higher accuracy are given a greater weight in the global model aggregation. This intelligent weighting can lead to global models that perform as well as, or even better than, the average of local models.
  • Low-Cost Hardware Feasibility: The study demonstrates that this sophisticated platform can be effectively implemented using low-cost, low-end hardware like Raspberry Pi, making it more accessible for real-world deployment.

Key Results

  • The implemented federated learning system, using a 1D CNN for HAR, achieved a testing accuracy of 90.43% for individual local models.
  • The confusion matrix of the global model, aggregated from 10 local models, showed strong performance in classifying human activities.
  • Figure 6 demonstrates that as the number of clients (fog nodes) increases, the global model's accuracy consistently keeps up with and eventually surpasses the average accuracy of the local models, peaking at 91.75%. This indicates the effectiveness of the proposed scaling scheme and the platform's ability to maintain or improve performance with more data.
  • The platform demonstrated effectiveness even when using low-cost devices (Raspberry Pi) as fog and cloud servers, highlighting its scalability and efficiency under resource constraints.
  • Qualitative analysis indicated strong privacy preservation (raw data stays local), security (blockchain access control), and adaptability (fog architecture handles device diversity).

Limitations and Future Work

  • Limitations:
    • The study focused on a specific HAR dataset and a 1D CNN, and the authors acknowledge that the performance might vary with different datasets and classifiers.
  • While the platform's feasibility was shown with low-cost hardware, the impact of processing speed on overall performance was not the primary focus.
  • The current implementation does not explicitly measure propagation delay or power consumption, which are important for real-world remote healthcare applications.
  • Future Work:
    • Investigate other performance metrics, such as propagation delay between fog and cloud, and power consumption of servers.
  • Explore the platform's performance with more diverse neural network designs and datasets.
  • Further optimize the design for portability and power efficiency to cater to remote monitoring applications.
  • Test the platform with more than one neural network design and dataset to emphasize its modularity and performance under different configurations.

Summary updated

Abstract

Over the years, the popularity and usage of wearable Internet of Things (IoT) devices in several healthcare services are increased. Among the services that benefit from the usage of such devices is predictive analysis, which can improve early diagnosis in e-health. However, due to the limitations of wearable IoT devices, challenges in data privacy, service integrity, and network structure adaptability arose. To address these concerns, we propose a platform using federated learning and private blockchain technology within a fog-IoT network. These technologies have privacy-preserving features securing data within the network. We utilized the fog-IoT network’s distributive structure to create an adaptive network for wearable IoT devices. We designed a testbed to examine the proposed platform’s ability to preserve the integrity of a classifier. According to experimental results, the introduced implementation can effectively preserve a patient’s privacy and a predictive service’s integrity. We further investigated the contributions of other technologies to the security and adaptability of the IoT network. Overall, we proved the feasibility of our platform in addressing significant security and privacy challenges of wearable IoT devices in predictive healthcare through analysis, simulation, and experimentation.
BibTeX
@article{baucas2023federated,
  author = {Marc Jayson Baucas and Petros Spachos and Konstantinos N. Plataniotis},
  title = {{Federated Learning and Blockchain-Enabled Fog-IoT Platform for Wearables in Predictive Healthcare}},
  journal = {IEEE Transactions on Computational Social Systems},
  year = {2023},
  url = {https://openreview.net/forum?id=s4nkaxRxyK},
}
Figure