Detecting online auction fraud

By Everett Dorma Posted: July 13, 2015 6:00 a.m.

Dr. Sadaoui reviewing an online auction in progress
Dr. Sadaoui reviewing an online auction in progress (Photo: U of R Photography).

You’re bidding on an item in an online auction. How do you know you’re not being a victim of fraud?

Computer Science professor Dr. Samira Sadaoui and Masters students Xuegang Wang and Swati Ganguly are developing a system to detect and react to shill activity in online auctions.
Auction fraud, which represents the majority of all Internet frauds, is on the rise, and in some situations the innocent bidders are not even aware that they’ve been defrauded.
Shill bidding has been recognized as one of the most dominant cheating activities in online auctions and also one of the hardest to detect since there is usually no evidence of its occurrence, unlike pre- and post-auction frauds. A shill is a person who poses as an ordinary bidder with the intent of increasing the price of the auctioned item.

“Several studies of online auctions have shown that shills are operating at popular auction houses,” says Dr. Sadaoui. “However, these research projects have only been completed after the auction is over, which is too late as the honest bidders have already been cheated.  What we’re working on is a method to analyze and detect fraudulent activity in real time as the auction is underway so we can stop the fraudsters from succeeding.”

research team examining online auction fraud
(L-R) Researchers Xuegang Wang, Dr. Samira Sadaoui and and Swati Ganguly are developing a system to detect and react to fraudulent activity in online auctions (Photo: U of R Photography)

Various shill bidding patterns have been identified and placed into several classes including:

  • Security related, a shill bidder has alternate identities by using different accounts and IP addresses.
  • Collusive behaviour, a shill participates exclusively in auctions held by some particular sellers, colluding bidders work together to inflate the price, sellers place bids on each other’s auctions, or users who live in a proximity area collude.
  • Competitive shilling, a shill aggressively increases the price, or bids more often;
  • Buy back shilling, the seller or his accomplice wins the auction when the price is considered too low in order to re-sell the item again.
“Through various statistical analysis and machine learning mechanisms we can detect these shill patterns as the auction is progressing,” says Wang.  “Once shill activity has been detected the auctioneer can, depending on the confidence level, notify the suspected shill bidder that they have been flagged as a possible shill and warn the other bidders of the potential presence of a shill, or they can cancel the auction for that item.”

By monitoring individual bidders over several auctions they can be classified as normal, suspicious and fraudulent. Auction houses can then suspend the accounts of bidders consistently identified as fraudulent to prevent them from participating in future auctions.   

“By detecting fraudulent shill activity in real time we can reduce participants’ risks and improve the reputation of online auction houses,” says Sadaoui. “Future development of our online auction fraud detection system includes implementation and evaluation of the system on very large data sets of auctions and users and using additional detection mechanisms to improve the system’s monitoring performance.”

This research project is supported by a grant from the Natural Sciences and Engineering Research Council. Research impact is identified as a strategic priority in the University of Regina’s new strategic plan.