Cost-Efficient Shuffling and Regrouping Based Defense for Federated Learning