A Study on Delivery Evaluation under Asymmetric Information in the Mail-order Industry

This paper presents investigating the fraud transaction detection in the mail order industry. These kinds of detection made intensively but the outcome of the research was not shared among the industry. As the B2C industry expands their market size, the fraud transactions increase in number. As a matter of course, this phenomenon is not only continuing but cleverly. One of the conclusive factors for this phenomenon is payment method. That is, the deferred payment method is primarily employed in Japan. The conventional primary indicator for the fraud detection is the ordered time-based information. They are the shipping address, the recipient name, and the payment method. Since conventional detecting method for the fraud depends on some heuristic knowledge, their market size enlargement makes hard to detect fraud transaction. For this background, this paper is presented investigating for comparing algorithms with the actual transaction data gathered from the mail-order industry in Japan. The comparison of weaker learner algorithms is made. The analytical results suggest Random forest is more accurate than XGBoost not only AUC score but parameter tuning costs. This result will make it use for the decision support knowledge for screening customer at the order received phase in the mail order industry.