Addressing Fairness in Classification with a Model-Agnostic Multi-Objective Algorithm