Stanford University - Department of Statistics
Classification problems get harder as the number of classes increase. For example, the null error rate is 1-1/K where K is the number of classes. Classes can mask each other in structured parametric settings. Some recent proposals can be viewed as "Plug in Classifier" techniques or PACTS. One solves a multi-class problem by solving many smaller problems, involving a much smaller number of classes. These are then combined in an appropriate way to perform the final classification.