Fusing one-class and two-class classification – A case study on the detection of pepper fraud