Integrating White and Black Box Techniques for Interpretable Machine Learning