A Coverage-based Approach to Nondiscrimination-aware Data Transformation

The development of technological solutions satisfying nondiscriminatory requirements is one the main current challenges for data processing. Back-end operators preparing, i.e., extracting and transforming, play a relevant role w.r.t. nondiscrimination, since they can introduce bias with an impact on entire life-cycle. In this article, we focus back-end transformations , defined in terms Select-Project-Join queries, coverage . Coverage aims at guaranteeing that input, or training, dataset includes enough examples each (protected) category interest, thus increasing diversity aim limiting introduction during next analytical steps. article proposes approach to automatically rewrite transformation result violates constraints, into “closest” query constraints. approximate relies sample-based cardinality estimation, it introduces trade-off between accuracy efficiency. efficiency effectiveness are experimentally validated synthetic real data.