'This paper argues that comparative studies of supervised classification often fail to take into account important aspects of real problems, so that the apparent superiority of more sophisticated methods may be something of an illusion.' (via adulau)