Differentially private histogram publishing through Fractal dimension for dynamic datasets

Abstract

Differential privacy is a prevalent research area that has been widely explored in the data release and analysis in recent decades. So far, we have focused on statistics that are derived from a static dataset. However, numerous applications require continuous publication of statistics, such as real-time disease outbreak discovery. With this consideration, a differentially private histogram publishing method with fractal dimension mining technology was proposed. The work aimed at hiding sensitive information while publishing, improving data utility and processing efficiency for multi-dimensional data. We used the fractal dimension to cluster datasets and counted values of each class. Through adding another algorithm to release the final histogram with Laplace noise, differential privacy is achieved. Extensive experiments with several real datasets confirm that our proposal achieves better privacy protection for dynamic datasets.