Imagine you are an expert hired to make prediction on one's income. Given some information about a person, your task is to predict if the person's annual income level is greater than 50K$ or not. The task, however, is not as simple as you thought initially. Rather than purely maximizing prediction accuracy, your employer has an extra requirement on the fairness of your predictions. Specifically, your prediction must follow a relaxed version of demographic parity for the binary sensitive attribute "gender". For details, check out the
Evaluation
section