AI-UBA: An Intelligent Multi-Objective Framework for Adaptive Privacy-Utility Optimization in Data Publishing

Abstract

Balancing privacy protection with analytical utility remains a fundamental challenge in privacy-preserving data publishing. Building on the Utility-Based Anonymization (UBA) framework, which adapts anonymization according to data sensitivity and intended utility, this paper proposes AI-UBA, an intelligent multi-objective framework that dynamically optimizes privacy, utility, and information loss. AI-UBA integrates machine-learning-based sensitivity and disclosure-risk estimation with adaptive selection of privacy-preserving mechanisms and Pareto-based multi-objective optimization. Unlike static anonymization approaches, the framework considers attribute correlations, contextual disclosure risk, and downstream analytical requirements when determining the appropriate protection strategy. Experimental evaluations on benchmark datasets is expected to demonstrate that AI-UBA can reduce information loss and disclosure risk while preserving higher predictive and statistical utility compared with conventional k-anonymity, differential privacy, and fixed-parameter anonymization baselines. The principal contribution is a unified, explainable, and task-aware framework that transforms UBA from a rule-based approach into an adaptive AI-driven privacy-utility optimization system, enabling safer and more useful data publishing across heterogeneous analytical scenarios.