We consider the risk, or probability of error, of the classifier produced by AdaBoost, and in particular the stopping strategy to be used to ensure universal consistency. (A classification method is u
