پیدا کردن سوزن در انبار کاه: رتبه بندی مبتنی بر ریسک لیست های محصولات در سایت های حراج آنلاین برای پیش بینی کلاهبرداری عدم تحویل

Non-delivery fraud is a recurring problem at online auction sites: false sellers that list nonexistent products just to receive payments and afterwards disappear, possibly repeating the swindle with another identity. In our work we identified a set of publicly available features related to listings, sellers and product categories, and built a machine learning system for fraud prediction taking into account the high class imbalance of real data and the need to control the false positives rate due to commercial reasons. We tested the proposed system with data collected from a major Brazilian online auction site, obtaining good results on the identification of fraudsters before they strike, even when they had no previous historical information. We also evaluated the contribution of category-related features to fraud detection. Finally, we compared the learning algorithm used (boosted trees) with other state-of-the-art methods.

Online auction sites like eBay offer unprecedented business possibilities for sellers and buyers through the creation of virtual marketplaces of global reach. Criminals also realized the opportunities opened by such virtual marketplaces. Among the several types of fraudulent behavior that take place in online auction sites, the most frequent one is non-delivery fraud (Gavish and Tucci, 2008 and Gregg and Scott, 2008): fake sellers list nonexistent products for sale, receive payments and disappear, possibly reentering the market with a different identity. According to the Internet Crime and Complaint Center (Internet Crime & Complaint Center, 2011), non-delivery fraud is the fourth most reported Internet crime. The challenge faced by site operators is to identify fraudsters before they strike, in order to avoid losses due to unpaid taxes, insurance, badmouthing etc. ( Chang & Chang, 2011). In other words, for a given product listing they need to predict whether or not it will end up being a fraud case. Since online auction sites are huge information systems and all transactions are carried over electronically, a natural approach to the fraud prediction problem is to use machine learning techniques. In this paper we will present a system for predicting non-delivery fraud that takes as input a set of product listings of an online auction site and outputs for each listing a fraud score, which can be used to analyze listings in decreasing order of risk. It also chooses a risk threshold so as to satisfy the user constraint on the rate of false positives. The proposed system uses a combination of features from product, seller and category, and, unlike other systems in the literature, it depends neither on historical data nor on social networks about the sellers in question, which is an advantage when dealing with fraudsters without reputation. The features we used can be extracted from the public web pages of online auction sites, which means that our system could be implemented by a third party, without the need of internal information. We evaluated the proposed system using data collected from a major Brazilian online auction site. In Section 2 we will present the context for our research; in Section 3 we will describe the dataset used to validate our approach and will present the selected features; in Section 4 we will explain our proposed system for predicting non-delivery fraud; in Section 5 we will present the experimental results, and in Section 6 we will discuss them.