A K-Farthest-Neighbor-based approach for support vector data descriptio