A Ranking-based Cascade Approach for Unbalanced Data