Transforming Feature Space to Interpret Machine Learning Models