A Weighted Privacy Mechanism under Differential Privacy
D Hemkumar, Pvn Prashanth · 2024
As digital technology advances, electronic databases are amassing an ever-growing volume of personal data. Even with the removal of explicit identifiers, such as names and addresses, the remaining data fields can still be leveraged to create distinct individual profiles, posing a significant threat to privacy. Therefore, safeguarding the privacy of individuals within datasets is of utmost importance. One highly effective technique for preserving privacy is differential privacy. Existing literature has employed various methods, such as Laplace and output perturbation, to achieve differential privacy. However, the accuracy of these mechanisms varies depending on the specific queries and the chosen value of є. Consequently, there exists a necessity for a differential privacy mechanism that can maintain accuracy across a wide range of queries. In this paper, we propose a method within the differential privacy framework known as the “weighted mechanism’’, which integrates elements from both Laplace and perturbation mechanisms. This approach involves releasing outputs with the addition of random noise from a defined distribution. To evaluate its effectiveness, we conduct a series of experiments comparing the accuracy of established differential privacy methods with the proposed weighted mechanism across various types of queries.