Real-Time Stream Data Anonymization via Dynamic Reconfiguration with l-Diversity-Enhanced SUHDSA.

Area of Science:

• Data Privacy and Security
• Information Systems
• Computer Science

Background:

• Traditional k-anonymity is insufficient for protecting against attribute disclosure with skewed sensitive attributes.
• High-throughput data streams require real-time anonymization techniques that respect strict delay budgets (β).

Purpose of the Study:

• To develop and evaluate a real-time anonymization strategy that jointly enforces k-anonymity and l-diversity for data streams.
• To minimize information loss and distortion while maintaining privacy guarantees under delay constraints.

Main Methods:

• Implemented a delay-aware Monitor-Trigger-Repair controller for real-time anonymization.
• Employed a weighted objective function (λΔIL + (1 - λ)ΔRT) to balance information loss (IL) and real-time constraints (RT).
• Utilized overhead bounding mechanisms (neighbor cap 'c' and growth cap 'γ') and evaluated an adaptive controller.

Main Results:

• Identified operating regions where enhanced privacy (k-anonymity and l-diversity) did not significantly increase data distortion, especially with moderate-to-high k and sufficient delay budgets (β).
• Demonstrated improved privacy metrics, including l-satisfaction rate and entropy.
• Showcased that parameters like λ, c, and γ offer interpretable trade-offs between latency and distortion, suppressing reconfiguration and tail latency.

Conclusions:

• Jointly enforcing k-anonymity and l-diversity in real-time data streams is feasible and can outperform k-anonymity alone.
• The proposed controller effectively balances privacy, distortion, and latency, offering practical solutions for high-throughput data anonymization.
• Parameter tuning provides flexibility for optimizing anonymization strategies based on specific application requirements.