Better Safe than Sorry - Implementing Reliable Health Data Anonymization.

Area of Science:

• Biomedical Informatics
• Computer Science
• Data Privacy

Background:

• Modern biomedical research generates large datasets, necessitating health data sharing and reuse.
• Data sharing raises significant privacy concerns, making robust anonymization techniques essential.
• Current anonymization methods often rely on floating-point approximations, potentially compromising formal privacy guarantees.

Purpose of the Study:

• To address the challenge of reliability in data anonymization tools.
• To investigate the impact of numerical approximations on privacy guarantees.
• To develop and evaluate a reliable computing framework for data anonymization.

Main Methods:

• Developed a reliable computing framework utilizing fractional and interval arithmetic.
• Implemented and tested the framework for data anonymization applications.
• Conducted extensive evaluations to assess reliability, execution time, and data utility.

Main Results:

• Demonstrated the practicality of reliable data anonymization.
• Showcased that the proposed framework improves the reliability of privacy implementations.
• Evaluations indicated minor impacts on execution times and data utility.

Conclusions:

• Reliable computing frameworks are essential for trustworthy data anonymization in practice.
• Fractional and interval arithmetic offer a viable solution for enhancing the accuracy of privacy models.
• The developed approach enables practical, reliable health data anonymization without significant performance or utility trade-offs.