Improving statistical learning methods via features selection without replacement sampling and random projection