Non-Interactive Correlation Differential Privacy for Healthcare Data

Abstract

The current differential privacy protection methods for healthcare datasets have achieved some results, but there are still some problems to be solved : (1) Most of the existing methods consider the privacy leakage caused by record correlation, and there are few studies on the privacy leakage caused by feature correlation. (2) Existing research also uses feature extraction methods when dealing with feature correlation, but the set extracted by this method is not the maximal feature set, which reduces the availability to a certain extent. In view of the above considerations, this thesis proposes a noninteractive correlation differential privacy protection method for healthcare data. First, In terms of feature correlation, an undirected graph is constructed with features as nodes, and the concept of feature sensitivity is proposed. Secondly, The Bron-Kerbosch algorithm is used to solve the maximal independent set, which solves the problem of information leakage caused by slow processing speed and feature correlation caused by multiple features. Finally, Experiments on two different datasets verify the effectiveness of the proposed differential privacy protection method.