Disparate Impact

Disparate Impact Is a fairness measure to assess how a given predictor behaves when applied in groups having different sensitive attributes. Disparate impact helps answering the question: how is the prevalence of subpopulation 1 compared with subpopulation 2 ?. Mathematically is given by the above formula.

Click here to see a video that explains in detail statistical parity.

In Kafkanator, you have an easy method to quickly compute the most common fairness metrics and to display them in a matrix. Download here a notebook to do it.