Developing a Fidelity Evaluation Approach for Interpretable Machine Learning