Predicting Disease Risks from Highly Unbalanced Data using Random Forest