A simple tool analyzing a dataset and computing for each value of each sensitive attribute given by the user as input:
- the Local Discrimination, i.e., the distance from a statistical parity condition. For non-binary sensitive attributes, a one-vs-all scenario is supposed;
- the Global Discrimination, i.e., the percentage of records to be modified to get a "fair" dataset.
Based on GenFair's Discrimination Test.
A library will be provided soon.