• bias identification
• fairness
• subgroup discovery

The thesis concerns the creation of a framework that allows to provide support for the discovery of unequally treated subgroups in a machine learning model, or that appear more discriminated in a dataset. This was done through the use of subgroup discovery algorithms combined with the most widely used fairness metrics. The developed framework was used for the analysis of the UCI adult dataset, which reports the data of about 48,000 individuals, together with a label indicating whether they earn more or less than 50k per year. The framework made it possible to bring to light the subgroups of the dataset within which there is a high disparity between men and women.