Class Restricted Clustering and Micro-Perturbation for Data Privacy.

Area of Science:

• Computer Science
• Data Privacy
• Information Security

Background:

• Organizations extensively use information technology, leading to significant privacy concerns regarding personal data collection and sharing.
• Clustering-based data masking techniques are vital for privacy-preserving data sharing and analytics.
• Traditional methods inadequately address disclosure risks for categorical attributes.

Purpose of the Study:

• To develop a novel clustering-based data masking approach that mitigates disclosure risks for categorical attributes.
• To enhance privacy-preserving data sharing by ensuring confidential attributes are well-distributed within clusters.
• To preserve key statistical properties (variance, covariance) often lost in traditional masking methods.

Main Methods:

• A Minimum Spanning Tree (MST) technique is employed for data clustering.
• Two risk-utility tradeoff measures are integrated into the MST growing and pruning stages.
• A novel cluster-level micro-perturbation method is proposed for data masking.

Main Results:

• The proposed method clusters data based on similar non-confidential attributes while ensuring well-distributed confidential attributes.
• The micro-perturbation method generates unbiased estimates of the original mean vector and covariance matrix.
• Experimental studies on real-world datasets validate the approach's effectiveness.

Conclusions:

• The novel clustering and micro-perturbation approach effectively balances privacy protection and data utility for categorical attributes.
• This method overcomes limitations of traditional techniques in preserving statistical properties.
• The findings contribute to more robust privacy-preserving data sharing and analytics solutions.