Differential privacy fuzzy C-means clustering algorithm based on gaussian kernel function.

Area of Science:

• Data Mining
• Privacy-Preserving Machine Learning
• Clustering Algorithms

Background:

• Fuzzy C-means (FCM) is a common clustering algorithm in data mining.
• FCM is vulnerable to privacy leaks due to sensitive dataset information.
• Differential privacy methods can protect user privacy but often reduce data availability and accuracy.

Purpose of the Study:

• To enhance the accuracy and effectiveness of fuzzy C-means clustering with differential privacy.
• To address the issue of reduced algorithm accuracy caused by random initialization in FCM.
• To develop a privacy-preserving clustering method that maintains high data utility.

Main Methods:

• Utilized the maximum distance method for initial center point determination in FCM.
• Employed Gaussian values of cluster centers to calculate privacy budget allocation ratios.
• Integrated Laplace noise addition for robust differential privacy protection.
• Evaluated clustering accuracy and effectiveness against baseline algorithms.

Main Results:

• The proposed algorithm demonstrated superior clustering accuracy compared to baseline methods.
• The enhanced FCM algorithm showed improved effectiveness under identical privacy protection levels.
• The method successfully balanced privacy preservation with data utility.

Conclusions:

• The novel approach effectively protects user privacy during fuzzy C-means clustering.
• The maximum distance initialization and Gaussian-based budget allocation improve algorithm performance.
• This method offers a promising solution for privacy-preserving data mining applications.