How to ensure interpretability in machine learning models