FAItH: Federated Analytics and Integrated Differential Privacy with Clustering for Healthcare Monitoring.

Area of Science:

• Health Informatics
• Data Science
• Privacy-Preserving Technologies

Background:

• Physical activity monitoring is vital for patient health, chronic disease management, and rehabilitation.
• Wearable devices collect crucial data, but privacy regulations (e.g., GDPR) and security concerns limit collaborative healthcare analysis.
• Federated analytics (FA) allows insights without data sharing, yet research often prioritizes data protection over actionable outcomes.

Purpose of the Study:

• To address the gap in analyzing privacy-preserved data for patient monitoring and healthcare interventions.
• To propose FAItH, a dual-stage solution integrating privacy-preserving techniques with federated analytics.
• To evaluate the trade-off between privacy and utility in analyzing aggregated patient activity data.

Main Methods:

• Implemented FAItH, integrating Laplace, Gaussian, Exponential, and Locally Differentially Private (LDP) noise with statistical functions (mean, variance, quantile).
• Employed feature-specific scaling to optimize the privacy-utility balance for sensitive and non-sensitive features.
• Utilized clustering on privacy-preserved, aggregated data to identify patient activity patterns.

Main Results:

• FAItH demonstrated that privacy-preserving configurations achieved clustering utility comparable to non-DP methods.
• The proposed solution outperformed existing privacy-preserving clustering algorithms in utility.
• Feature-specific scaling effectively managed the privacy-utility trade-off.

Conclusions:

• Federated analytics with differential privacy is a viable solution for secure collaborative healthcare analysis.
• FAItH enables the extraction of meaningful insights from patient activity data without compromising privacy.
• The approach supports effective patient monitoring and healthcare interventions in a privacy-conscious manner.