Learning decision trees from random examples