It would also be interesting to investigate the effect of round robin binarization on minority classes, in particular in problems where several large classes appear next to a few small classes. We think that the fact that separate classifiers are trained to discriminate the small classes from each other (and not only from all remaining examples as would be the case for unordered binarization or for treating the multi-class problem as a whole) may help to improve the focus in the case of imbalanced class distributions. On the other hand, if the base learner tends to prefer large classes, one dominant large class will tend to win against all minority classes and will be more frequently predicted. The evidence from Table 4 seems to confirm this: it is primarily sets with skewed class distributions where round robin classification does not perform well (consider, e.g., the three thyroid datasets). However, this evidence is certainly not conclusive and we believe that a closer investigation of this issue is a rewarding topic for future research.