Explainable AI Evaluation: A Top-Down Approach for Selecting Optimal Explanations for Black Box Models