A Probabilistic Approach to Mitigate Composition Attacks on Privacy in Non-Coordinated Environments.

Area of Science:

• Computer Science
• Information Security
• Data Privacy

Background:

• Organizations share data for business and legal reasons, but this can expose individuals' confidential information.
• Composition attacks, using quasi-identifying attributes, compromise privacy in existing models like k-anonymity.
• Coordination between organizations can prevent attacks but is not always feasible.

Purpose of the Study:

• To introduce a novel probabilistic model, (d, α)-linkable, to mitigate composition attacks.
• To provide a privacy solution that does not require prior coordination between data publishers.
• To enhance individual privacy while maintaining data utility.

Main Methods:

• Developed a probabilistic model, (d, α)-linkable, ensuring 'd' confidential values link to a quasi-identifying group with probability 'α'.
• Implemented the model through an efficient extension of k-anonymization.
• Conducted extensive experiments to evaluate the model's effectiveness and utility preservation.

Main Results:

• The (d, α)-linkable model significantly reduces the likelihood of successful composition attacks.
• The proposed strategy preserves more data utility compared to alternative privacy models, including differential privacy.
• The method effectively mitigates privacy risks without necessitating inter-organizational data sharing coordination.

Conclusions:

• The (d, α)-linkable model offers a robust solution for privacy protection against composition attacks.
• This approach enhances data security and utility, making it a valuable tool for data publication.
• The model provides a practical alternative for safeguarding individual privacy in data sharing scenarios.