A Comparative Study of Faithfulness Metrics for Model Interpretability Methods