Federated Mining of Interesting Association Rules Over EHRs

Area of Science:

• Computer Science
• Data Mining
• Artificial Intelligence

Background:

• Federated learning enables collaborative model training without centralizing sensitive data.
• Existing federated learning techniques often lack transparency, hindering explainability and auditability.
• Discovering meaningful patterns, such as association rules, is crucial for data mining applications.

Purpose of the Study:

• To propose a novel federated data mining method for association rule discovery.
• To enhance the explainability and auditability of federated learning processes.
• To align federated data mining with FAIR data principles (Findability, Accessibility, Interoperability, Reusability).

Main Methods:

• Definition of 'interesting itemsets' within a federated context.
• Development of a specific algorithm to identify these itemsets across distributed data sources.
• Focus on methods that do not require raw data transfer between sources.

Main Results:

• A new federated data mining approach for association rule discovery.
• The proposed method addresses the explainability and auditability gap in federated learning.
• Demonstration of how the method supports data interoperability and reusability.

Conclusions:

• The developed federated data mining method offers a transparent and auditable way to discover association rules.
• This approach promotes data accessibility and aligns with FAIR data principles.
• It paves the way for more trustworthy and practical federated data mining applications.