Semi-local Time sensitive Anonymization of Clinical Data.

Area of Science:

• Data privacy and security
• Time-series analysis
• Information anonymization

Background:

• Protecting sensitive time-continuous data from linking and distribution attacks is crucial.
• Existing anonymization methods may not adequately preserve the temporal and value dimensions of data.
• Ensuring robust privacy guarantees like k-anonymity and t-closeness is essential for data utility.

Purpose of the Study:

• To propose a novel method for anonymizing time-continuous data.
• To preserve the relationship between time and value dimensions.
• To provide strong privacy guarantees against linking and distribution attacks.

Main Methods:

• Windowed Fréchet Splitting for time-axis segmentation to minimize information loss.
• Distribution Clustering for generating representative population distributions.
• Bucketization using Fréchet distance with implicit cost and t-closeness.
• Multiple redistribution phases and semi-local decisions for runtime optimization.

Main Results:

• Achieved low information loss and median relative error.
• Demonstrated effective t-closeness for privacy protection.
• Significantly reduced runtime through semi-local decision-making.

Conclusions:

• The proposed method effectively anonymizes time-continuous data while preserving its integrity.
• The approach offers robust protection against sophisticated data attacks.
• The method balances privacy preservation with data utility and computational efficiency.