Privacy preserving data visualizations.

Area of Science:

• Data Science
• Statistical Analysis
• Information Security

Background:

• Data visualizations are crucial for understanding complex datasets, revealing patterns and relationships hidden in raw data.
• Sharing sensitive data, common in medicine and social sciences, is restricted by privacy laws and ethical guidelines.
• Traditional data visualizations risk re-identification of individuals, leading to prohibitions on their use in data sharing.

Purpose of the Study:

• To propose and evaluate anonymization techniques for generating privacy-preserving data visualizations.
• To enable the display of descriptive plots and diagnostic plots for research without compromising participant confidentiality.
• To maintain the statistical properties of the original data in the generated visualizations.

Main Methods:

• k-anonymization: reducing data granularity through suppression and generalization.
• Deterministic approach: replacing observations with centroids of k-nearest neighbors.
• Probabilistic approach: perturbing attributes with random noise.

Main Results:

• Demonstrated the generation of privacy-preserving visualizations using the proposed anonymization techniques.
• Applied methods to exploratory data analysis and inferential regression plot diagnostics.
• Evaluated the strengths and limitations of each anonymization approach for visualization.

Conclusions:

• Anonymization techniques offer a viable solution for creating informative yet privacy-preserving data visualizations.
• These methods facilitate the ethical sharing and analysis of sensitive data, overcoming disclosure restrictions.
• The proposed techniques balance data utility with robust confidentiality protection for research.