Privacy-Preserving Data Sharing Using Differential Privacy in Healthcare

Abstract

Abstract: This study evaluates the effectiveness of differential privacy (DP) techniques for privacy-preserving data sharing in healthcare environments. Healthcare data is highly sensitive, making data sharing both a valuable and risky endeavor. Traditional anonymization techniques fail to provide sufficient guarantees against re-identification attacks, particularly in high-dimensional datasets. Differential privacy offers mathematically rigorous privacy guarantees while enabling data utility. Using real-world healthcare datasets, the proposed approach evaluates the trade-off between privacy budget (ε) and model accuracy when training logistic regression and decision tree classifiers. A Laplace mechanism is applied to numerical attributes, and post-processing techniques are introduced to balance accuracy and privacy. The results demonstrate that meaningful model performance can be maintained while reducing privacy risks through carefully selected privacy budgets. Regression analysis and predictive evaluation confirm the viability of DP in healthcare data sharing. This research provides empirical evidence and best practices for deploying DP-based data sharing frameworks in clinical analytics. Keywords Differential Privacy, Healthcare Data, Data Sharing, Privacy-Preserving, Laplace Mechanism, Regression Analysis