PrivacyGuard: Differentially Private Machine Learning Against AI-Driven Background Knowledge Attacks

View at publisher Summarise with VIRAT
Authors: Shoaib Ullah, Razaullah Khan, Madiha Haider Syed, Adeel Anjum
Cited by: 0
References: 33

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Preserving an individual privacy in the Artificial Intelligence(AI)era is a pressing challenge, as traditional anonymization techniques often fall short againstAI-driven threats. WhileAIhas enabled transformative advancements across sectors such as healthcare, finance, aviation, smart cities, and the Internet of Things (IoT), it also introduces new privacy challenges. In particular,AI-driven background knowledge attacks occur when adversaries combine auxiliary information from external sources such as public records, social media, or previously leaked datasets with advancedAItechniques to re-identify individuals or infer sensitive details from anonymized datasets, rendering conventional defenses less effective. In this study, we presentPrivacyGuard, an automated framework that integrates machine learning with traditional anonymization to address these risks.PrivacyGuardemploys K-Nearest Neighbor and Random Forest classifiers to identify and evaluate privacy risk of each data tuple inAI-driven attack scenarios. Following risk assessment, Laplace-based differential privacy is applied to mitigate identified risks, while fuzzy logic forms equivalence groups that reflect varying risk levels. Experimental validation on the 1:M Adult dataset (an individual having multiple records) demonstrates thatPrivacyGuardachieves 96.47% accuracy and mitigates approximately 42.21% of the identified risks, confirming its effectiveness in balancing privacy and data utility against emergingAI-driven threats.

Fields of Study & Subjects

Privacy-Preserving Technologies in Data Privacy Security and Data Protection Internet Traffic Analysis and Secure E-voting

Publication details

Journal / source
IEEE Transactions on Dependable and Secure Computing
Volume
23
Issue
4
Pages
7990–8001
Publication date
31 March 2026
Language
EN