A Fraudster in a Haystack: Crafting a Classifier for Non-delivery Fraud Prediction at Online Auction Sites

Non-delivery fraud is a recurring problem at online auction sites: false sellers that list inexistent products just to receive payments and disappear, possibly repeating the swindle with another identity. The high transaction volume of these sites calls for the use of machine learning techniques in fraud prediction systems, at least for the identification of suspect sellers which deserve further expert analysis. In our work we identified a set of features related to listings, sellers and product categories, and built a system for fraud prediction taking into account the high class imbalance of real data, since fraud is a relatively rare event. The identified features are all based on publically accessible data, opening the possibility of developing fraud prediction systems independent of site operators. We tested the proposed system with data collected from a major online auction site, obtaining encouraging results on identification of fraudsters before they strike, while keeping the number of false positives low.