In their original exposition on ADTrees, Freund and Mason [7] note that
because alternating trees can be defined as a sum of simple base rules it is
a simple matter to apply any boosting algorithm to the problem of inducing
ADTrees. For the multiclass setting one possible candidate is AdaBoost.MH.
In this paper we also explore and compare two other solutions. The first is to
adapt the original two-class ADTree algorithm to the multiclass setting using a
variety of wrapper methods. The second is to use the multiclass LogitBoost [10]
procedure as the underlying boosting algorithm. This algorithm is a natural
choice as it is directly applicable to multiclass problems.