On the Applicability of ML Fairness Notions