The Research Of Credit Card Anti-Fraud System Base On Neural Network

In recent years, credit card risks have come into focus of credit card business area, as domestic banks have been offering various types of credit cards. The fraud risk as one of the credit card risks brings banks and customers with the most significant losses. Foreign credit card industry started earlier, and now they have owned mature technologies and related products on anti-fraud, but these technologies and products are based on some specific foundations which are not available in domestic banks. Therefore, in the current economic conditions of domestic banks, how to develop practical anti-fraud system to control the risk of credit card fraud and ensure the sound development of credit card industry has become very important.The existing anti-fraud models, anti-fraud systems and products at home and abroad are presented in-depth, and the analysis of the obstacles that foreign anti-fraud products encountered in the implementation and promotion at home, and the shortcomings in directly applying the existing anti-fraud products to credit anti-fraud. Base on this analysis and neural network technology, a model of credit card anti-fraud suitable for domestic commercial banks is proposed and its details of system processes and operating mechanism is analyzed. Finally, the model shows great rationality when it is used in one domestic bank for analyzing the credit card transaction in one period.Firstly, the concept and classification of credit card risks are introduced, as well as the importance of fraud risk. Then, the implemention style of two popular anti-fraud models are presented, whose advantages and disadvantages are compared. On this basis, we sum up the barriers forgein anti-fraud products encountered when promoting to domestic banks and its own flaws in applying to anti-fraud area, propes a feasible credit card anti-fraud model system suitable for domestic commercial banks based on neutral network.This model system consists four modules:data preprocessing module, neural network module, output module and tracking module. The modules consistent with system flow designment works compatibly with each other. In addition, the BP neutral network algorithm is improved, thus avoid overfitting without losing its model prediction ability. This anti-fraud model system features a modular design. The structure of the system design and module design are highly transparent. With the popular anti-fraud technology, neural network technology, this system requires little on banks' existing system. So it is very suitable for domestic commercial banks considering both development cost and maitance.