Automated k-Anonymization and l-Diversity for Shared Data Privacy

Analyzing data is a cost-intensive process, particularly for organizations lacking the necessary in-house human and computational capital. Data analytics outsourcing offers a cost-effective solution, but data sensitivity and query response time requirements, make data protection a necessary pre-processing step. For performance and privacy reasons, anonymization is preferred over encryption. Yet, manual anonymization is time-intensive and error-prone. Automated anonymization is a better alternative but requires satisfying the conflicting objectives of utility and privacy. In this paper, we present an automated anonymization scheme that extends the standard k-anonymization and l-diversity algorithms to satisfy the dual objectives of data utility and privacy. We use a multi-objective optimization scheme that employs a weighting mechanism, to minimise information loss and maximize privacy. Our results show that automating l-diversity results in an added average information loss of 7% over automated k-anonymization, but in a diversity of between 9-14% in comparison to 10-30% in k-anonymised datasets. The lesson that emerges is that automated l-diversity offers better privacy than k-anonymization and with negligible information loss.