Credit Card Fraud Detection Based On Fuzzy 2-norm Quadratic Surface Support Vector Machine

With the continuous development of the global economy and the further opening of financial markets,domestic and international credit card market has been expanding rapidly,credit card has not only become a very common tool of payment,but also become an important source of bank income.The credit card industry has brought great benefits to the banking industry as well as high risk.However,China's credit card started late,after 30 years of development,the credit card market has a certain scale,especially the past 10 years witnessed the explosive growth of credit cards,the huge credit card market has become the target of lawless elements.The rapid growth of credit card fraud risk,constantly refurbished means of fraud brought huge economic losses to the card issuers and cardholders and restricted the healthy development of the credit card industry.However,China has not yet established a sound credit system,and it also has not played a role in the credit industry.Domestic and foreign financial institutions are facing serious challenges.How to identify credit card fraud effectively,quickly and accurately has become the focus of financial institutions.Credit card fraud includes fraud application and fraud transaction.This paper analyzes and studies the credit card fraud risk from the application stage and the transaction stage,systematically sorts out the literature in the field of credit card fraud detection and then summarizes the challenges in credit card fraud detection and the insufficiency of existing literature.Credit card fraud data itself is characterized by imbalance,fraud data is often much less than the normal data.The existing literature usually use KNN classification,neural network and support vector machine(SVM)to solve the imbalance classification problem.KNN classification,neural networks are successful machine learning methods,which are widely used in various fields,but for credit card fraud,such a large data and the complex nonlinear relationship among data makes these methods face enormous challenges.Another commonly used classification method is support vector machine,the traditional support vector machine also has weakness,so on the basis of traditional support vector machine(SVM),this paper try to improve the SVM and to propose a kernel-free Quadratic Surface Support Vector Machine and Fuzzy two-norm Quadratic Surface Support Vector Machine(F2NQSSVM)for credit card fraud detection,then compare them to KNN,neural network and SVM.Finally,according to the example of fraud application data,we conduct a questionnaire survey of the banking industry to verify whether the results of the new model are consistent with the practical application and provide a new way for credit card fraud detection in financial institutions.The main tasks include:(1)The F2NQSSVM model are proposed.Firstly,we introduce the theory of traditional support vector machine and its simple application,and analyze the basic principle and related properties of SVM model.The main goal of the SVM model is to find a maximized interval hyper-plane that separates the two classes of training points as much as possible.In practice,we cannot find a hyper-plane which can completely separate all the points,the training data is often not linearly separable.The usual method is to map low-dimensional linearly indivisible points into high-dimensional space through a kernel function,which is linearly separable in high-dimensional space.However,there is no general selection criterion for kernel function selection and the selection of the kernel parameters directly determines the prediction results of the model.Then QSSVM model is proposed on this basis,and the QSSVM model is improved for the imbalance of credit card fraud data,which gives the higher penalty cost of fraud data.We also consider giving each training data a fuzzy membership and further proposed F2NQSSVM model.(2)A credit card fraud detection model based on QSSVM and F2NQSSVM model is established.Firstly,the raw data of application fraud and transaction fraud are preprocessed,we try to make feature transformation,feature selection and then determine the input of model.Then we use the QSSVM model and F2NQSSVM model to build credit card fraud detection model and compare them with KNN,neural network and SVM model,Which proves the validity and accuracy of the proposed new model.Finally,we use the application fraud data and take the example of Industrial and Commercial Bank of China(ICBC),China Everbright Bank(CEB),Dandong Bank practitioners to make a questionnaire survey to verify whether the result of the model is consistent with the real application,the practicality of the model,and finally provide some reasonable recommendations for the financial institutions.The main conclusions of this paper are as follows:1)For the data of credit card fraud and transaction fraud,the result of F2NQSSVM is the best,and the evaluation index F is the largest.The fuzzy method is effective for unbalanced classification and can significantly improve the effect of the model.Therefore,when dealing with unbalanced data,the method of fuzzy criteria can be used to improve the classification effect of the model,and the F2NQSSVM model can be effectively applied to credit card fraud detection.2)Through the research on fraud detection of credit card application data,the first eight important characteristics of the model should not only consider the applicant's housing condition,credit guarantee amount,the scope of employment years,the purpose of applying for a credit card,job skills,disposable The proportion of income,but also to consider the applicant's credit history and property status,therefore,for credit card applications,we need to continue to establish and improve the credit system,for credit card transactions,we need to timely monitor the large trading information and frequent trading information.The conclusions drawn above can provide reference and guidance for financial institutions to conduct credit card fraud detection.