Towards interpretable machine learning models for diagnosis aid: A case study on attention deficit/hyperactivity disorder