An Application Evaluation for Differentially Private Database Release Methods

Privacy violations involving de-anonymizing personal identifiable information (PII) due to data mining or breach is a corporate domain concern. Therefore, maintaining individual data anonymity has been challenging due to de-anonymization strategies employed to mine personal data. Counteracting de-anonymization data mining techniques have been deployed where anonymous data is cross-referenced with related data sources to de-anonymize the data source. Subsequently, privacy-enhancing technologies [PETs] have been suggested and deployed in academic and professional to prevent de-anonymization when extracting information from large datasets. Differential privacy (DP) is among the preferred privacy commitments among contemporary privacy models because it guarantees the protection of user-sensitive data. This research evaluates three differentially private database release methods, kernel mean embedding (KME), the learning theory (LT) approach to non-interactive database privacy, and the statistical framework for (SF) differential privacy. This evaluation indicates limitations and provides applicability suggestions for each method to maintain data anonymity and measures to overcome the challenges. Our study provides practitioners with guidelines on the DP selection based on the data nature.