Research About The Fraud Detection System Of Internet Finance Based On Data Mining

In 2015, Internet finance is surging, and draws investors’ attention. On the one hand, the Internet finance broke the traditional financial barriers so that borrowers can directly access to financing, and this innovative financial model has brought new opportunities for domestic financial development and China looks forward the infinite future of it. On the other hand, Internet financial-related adverse events are frequently exposed, and profit model of the Internet finance is being questioned. And the primary problem of Internet finance is strengthening supervision and reducing risk.The thesis unravels the wide range of Internet finance and divides it into three models:Firstly, the traditional Internet-based finance; Secondly, fund information service platform; Thirdly, the Internet-based platform to carry out the traditional financial services. The Internet banking in the trend of rapid development, is facing three major risks, credit risk, technical risk and regulatory risk. The risk of fraud as the main form of credit risk, is the primary problem of Internet finance to deal with, and therefore the construction of fraud detection is imminent.Based on data mining technology and preform research results, combined with discipline knowledge of statistics and machine learning, a set of reasonable fraud detection system of Internet finance is introduced. The fraud detection system is composed of three parts, statistical identification, pattern recognition, and artificial identification. Statistical identification is to collect, aggregate and analyze raw data by statistic methods; Model identification is the application machine learning algorithm; Artificial identification is the complement of the first two modules and check in the suspect fraud users again.In the theoretical framework of fraud detection system, the actual data is applied to analyzing and modeling. The raw data set contains the user’s consumption data, social data and credit data (fraud and normal). The results which are obtained by data mining shows that there are some relations among the users’ credit, consumptions and interaction, and the users’ social relationships have a significant impact on the users’ credit. Sampling part of the training set data and establishing neural network model, support vector machine model and random forest model. Using test data to validate those models, it shows that the neural network to has the optimal prediction results. Finally, the thesis summaries the fraud detection system, states the shortcomings of the study, and puts forward some reasonable options to develop the fraud detection of Internet finance.