Boosting for high-dimensional two-class prediction