Analyzing Auction Data for Anomalous Bidding

Online auctions as exemplified by sites such as ebay.com are responsible for very large volumes of transactions and monetary value. Their growth has also led to a growth in fraudulent activities in these markets. This paper studies transaction data from such auctions with the goal of using it to detect anomalous and potentially fraudulent bidding. To that end, it explores several approaches based on classification, clustering, and visualization. The quantitative results signal very high accuracy in classification but their promise is tempered by some limitations of the experimental dataset. Clustering and visualizations using self-organizing maps (SOMs) is found to be more effective for this data than clustering using more conventional methods such as k-means. In particular, the SOMs reveal several interesting relationships among the dataset’s attributes and their correlations to anomalous bidding.