The additional attributes appear to be useless. They seem to have had little effect on the performance, in some cases in fact leading to the error rate increasing. This means that perhaps we should ignore them and not use them at all, especially in the case of IBL1; they seemed to have had an improving effect on the results with C4.5.