Comparative Analysis of Federated Association Rules in a Simulated Environment for Medical Applications

Area of Science:

• Machine Learning
• Data Mining
• Medical Informatics

Background:

• Centralized databases pose privacy risks in healthcare data analysis.
• Federated Learning (FL) offers a decentralized approach to preserve data privacy.
• Increasing healthcare data and regulations necessitate privacy-preserving methods.

Purpose of the Study:

• To examine relevant association rule mining algorithms within unsupervised Federated Learning.
• To evaluate the performance of these algorithms in a simulated medical environment.
• To assess privacy preservation, execution time, and encryption/decryption efficiency in edge networks.

Main Methods:

• Utilized unsupervised Federated Learning (FL) for association rule mining.
• Simulated a decentralized environment with multiple nodes.
• Algorithms collaborated across nodes, sharing only computation results, not raw data.

Main Results:

• Demonstrated effective privacy preservation through encapsulated and encrypted raw data.
• Evaluated execution time and encryption/decryption efficiency in edge network scenarios.
• Confirmed FL's suitability for privacy-conscious medical data analysis.

Conclusions:

• Unsupervised Federated Learning is a viable paradigm for privacy-preserving association rule mining in healthcare.
• FL effectively safeguards sensitive patient data while enabling collaborative analysis.
• Further research can optimize FL for real-world medical applications.