Utility-Embraced Microaggregation for Machine Learning Applications

With access to vast amounts of data, privacy protection is more important than ever. Among various de-identification (anonymization) techniques, k-anonymous microaggregation has been widely studied since it enables us balance between confidentiality and data utility. Despite plenty methods in the sense reducing information loss and/or computational complexity, machine learning (ML) models using resulting aggregated face problem that they are not as effective expected. Motivated by fact ML can be heavily influenced distorted training (albeit slightly), we deliberate on performance terms only but also data utility. In this paper, propose Util-MA, a new utility-embraced framework for applications. Specifically, unlike prior studies apply techniques directly raw design unified potentially enhance utility while preserving k-anonymity through preprocessing steps including dimensionality reduction clustering. By real-world datasets, empirically demonstrate superiority Util-MA over benchmark classification accuracy. Moreover, investigate importance measuring key indicators (KPIs) clustering; clustering stage leads high when results substantially coincide with ground truth labels. We establish close relationship KPIs accuracies, which tends revealed there gain method observed. Our microaggregation-model-agnostic; thus, underlying appropriately chosen according one’s needs tasks.