Using clustering techniques to analyze fraudulent behavior changes in online auctions

Abstract

Schemed fraudsters often flip behavior in terms of circumstances change as camouflage for disguising malicious actions in online auctions. For instance, the fake transaction records interwoven with real trades are indistinguishable from legitimate transaction histories. The ways of fraudulent behavior changes formulate different types of tricks for swindling. To avoid trading with fraudsters, recognizing the types of fraudulent behavior changes in advance is helpful in choosing appropriate trading partners. Therefore, in order to distinguish the types of behavior changes from different fraudsters, clustering techniques were applied such as X-Means for grouping in characteristics. Afterwards, C4.5 decision trees were employed for inducing the rules of the labeled clusters. In this study, the real transaction data of 236 proven fraudsters was collected from Yahoo!Taiwan for testing. The experimental results demonstrate that the fraudsters are categorized into 4 natural groups and the vast majority of fraudster, 93% of fraudsters on average, follows certain default models to develop a scam. The findings of this study also make online auction early fraud detection possible.