Research On The B2B Platform Anti-fraud Problem

The fraud in B2B business platform has troubled the platform operators. The emergence of data mining technology brought a lot of help for the anti-fraud work in electric business platform. However, there are still many gaps and deficiencies in data mining technology area about how to solve these frauds. Firstly, there are many descriptive studies about online fraud, but there’s little actual solution to these problems in the real world. Secondly, we are still lack of real data for experiments. Thirdly, the researches in the past care little about the class imbalances and cost-sensitive issues.In order to effectively solve the problem of fraud on the B2B platform, this study choses an integrated classified algorithm-EasyEnsemble, There are two steps for EasyEnsemble to solve the class imbalance problems. The first step, by using the random sampling methods, negative categories can be divided into several subsets, and each subset balances with positive samples. Combining the subsets of negative categories with positive samples as training sets, we can get multiple classifiers. The second step is using Adaboost integration technology to combine the multiple classifiers.The data used in this article comes from the actual data of a large domestic B2B e-commerce company. At last I selected 2760 users’ information in 2011 and 2012. Among them, there are 2500 non-fraud records and 260 fraud items. Compared with common classification algorithms through experiments, we can see that EasyEnsemble algorithm is an effective measure to solve the problem of class imbalance. By using this algorithm, we can improve the accuracy of classification, reducing the misclassification ratio, as well as solving the problem of cost-sensitive. Then we use this algorithm to deal with the sample data. At last, we analysis the results through deep business view, and we provide support and suggestions for the anti-fraud problems on B2B platform.The major possible contributions of this article may be as below. Firstly, by using EasyEnsemble algorithm, we effectively improve the accuracy and the rate of misjudgment about the class imbalance problem. Besides, we also solve the cost-sensitive problems, proving a new thinking for further research. Secondly, by using the true data of B2B platform, the result of the classification can be more persuasive and convincing; making the research result contains more business sense. Thirdly, through the use of effective algorithms to deeply analysis the true data, we provide practical suggestions and important support for the anti-fraud issues on the domestic B2B platform.