Skip to Main Content
Online auction has become one of the most successful e-business models and created tremendous turnover rate for years. The large amount of monetary profit appeals fraudsters to step into online auctions. These fraudsters manipulate reputation systems to fabricate positive feedback score for attracting naive traders. Notwithstanding the vast majority of accounts are legitimate users, a tiny number of fraudsters threat the rest of legitimate trading participants severely. Thus, online auction fraud detection is treated as a kind of anomaly detection problem that is not as easy as expected. In general, a suspect was insufficiently judged as a fraudster without any actual victim as evidence. To reduce the risk of being defrauded, traders are necessary to have an assistant for choosing reliable trading partners. This paper proposed a mechanism based on instance-based learning to screen the transaction histories of trading partner candidates for estimating the risk of being defrauding. We downloaded real transaction histories from Yahoo!Taiwan for testing in this study. The experimental results show the recall rate of identifying potential fraudsters is 84%.