Performance Analysis of Various Differential Privacy Preserving Data Distortion Techniques using Privacy Class Utility Metric

Abstract

Statistical database security focuses on the protection of confidential individual data, stored in databases for statistical purposes. One of the techniques used for preserving statistical database privacy is noise addition. In this technique, in response to the queries, the statistical data provided as answers are only approximate rather than exact. In this background analysis of various techniques with heterogeneous data distortion is presented in this paper. An attempt is made, to study the effect of application of various statistical measures on the distorted data, and their impact on ensuring the privacy of the original data. Experimental results show that the proposed solution outperforms traditional differential privacy in terms of Statistical Metrics on a group of queries. The performance of heterogeneous data distortion is evaluated with three types of techniques namely homogeneous with differential privacy, heterogeneous with differential privacy and also sigmoid technique (Learning model) with differential privacy. It is observed that the sigmoid technique can successfully retain the utility of published data while preserving privacy.