Should Simplicity Be Always Preferred to Complexity in Supervised Machine Learning?