Building fast decision trees from large training sets