A differential privacy protecting K-means clustering algorithm based on contour coefficients.

Area of Science:

• Computer Science
• Data Privacy
• Machine Learning

Background:

• K-means clustering is a widely used algorithm for data analysis.
• Differential privacy techniques are essential for protecting sensitive information in data.
• Existing differential privacy methods for K-means can lead to inaccurate results with small privacy budgets.

Purpose of the Study:

• To improve the accuracy and availability of K-means clustering under differential privacy.
• To address the limitations of Laplace noise in standard differentially private K-means.
• To develop a privacy-preserving clustering algorithm suitable for large datasets.

Main Methods:

• Implemented a differential privacy K-means algorithm with Laplace noise addition to cluster centers.
• Introduced an improved algorithm utilizing contour coefficients for iterative clustering evaluation.
• Applied differential noise addition adaptively to different clusters based on their evaluation.
• Designed a MapReduce framework implementation for scalability with large datasets.

Main Results:

• The improved algorithm demonstrated enhanced availability of clustering results compared to standard methods.
• Individual privacy was maintained effectively while improving the utility of the clustering outcomes.
• The MapReduce implementation efficiently handled large-scale data processing.
• Performance analysis showed no significant increase in operating time.

Conclusions:

• The proposed enhanced differential privacy K-means algorithm effectively balances privacy and data utility.
• Contour coefficient-based noise adjustment is a viable strategy for improving differentially private clustering.
• The MapReduce framework enables scalable and private K-means clustering for big data applications.