A cost-effective method for early fraud detection in online auctions

Abstract

Online auction fraud has been one of the top 10 internet crime complaints for years. To protect legitimate traders from becoming victims of internet fraud, it is important to identify camouflaged fraudsters as early as possible. Users need an effective but efficient early fraud detection system to assist in making final trading decisions. To this end, a series of cost-effective measures are developed in this study. First, principal component analysis is applied to reduce the dimensionality of attributes set for describing the features of traders. In general, fewer attributes being applied implies less computation efforts. Afterwards, a late-profiling method is proposed to characterize fraudsters by behavior occurred in their last phase. By curtailing the transaction histories appropriately, the overall detection cost can be greatly reduced while retaining reasonable detection accuracy. Based on the above measures, a cost-effective procedure is then devised to perform lazy-downloading detection. To demonstrate the effectiveness of the proposed methods, real transaction data is collected from Yahoo!Taiwan for testing. The experimental results show that the detection accuracy of the late-profiling models built with the reduced attributes can be ranged from 91% to 95%. That is almost similar to the results of applying expensive detection methods in the previous research. In addition, while only part of transaction histories is available, the proposed methods can still maintain a high accuracy over 90%. For measured attribute set reduction, experimental results also show the principal components analysis is actually helpful in generating a concise but effective attribute set. In conclusion, the effectiveness of our work is clearly demonstrated by the experimental results.