Credit Card Fraud Prediction Model Based On Skip-GANormaly

With the continuous improvement of China’s economy,the consumption concept of the people has changed,not only to pursue a higher level of consumption,but also to explore a variety of consumption patterns.In recent years,credit card consumption as a pre-payment consumption model has been favored by people,the market scale is expanding,the business volume is growing.However,under the appearance of the booming credit card market,there are also hidden dangers,namely a series of risks caused by the use of credit cards,and credit card fraud is one of the main sources of risks.In this paper,how to prevent credit card fraud as the starting point of research,that is,to build an effective and efficient credit card fraud prediction model to reduce the loss and harm caused by credit card fraud.Usually,the establishment of credit card fraud prediction model needs to solve two problems: one is how to deal with the credit card transaction data samples with serious disequilibrium categories;The second is to choose what kind of classifier to meet the requirements of high prediction performance,namely the construction idea of data preprocessing + classifier selection.However,the prediction models of credit card fraud constructed in the past have some shortcomings,such as complicated feature construction,loss of under-sampled information,over-fitting of over-sampled model and poor prediction performance.To solve above problems,this paper constructed the Skip-GANormaly credit card fraud prediction model is mainly for character building and unbalanced data processing has been optimized,embodies in the following respects: First,stage of training model involves only the normal sample,normal data distribution characteristics of samples by learning to understand domain of population distribution,in the prediction stage,based on judging whether the current user data from the distribution of the user to detect whether the user is fraud;Second,The degree of manual intervention in the model construction process is low.Compared with the cumbersome data preprocessing of previous credit card fraud prediction models,the model in this paper only requires the basic data format processing;Third,In the field of credit card fraud prediction,Skip-GANormaly model is applicable to unbalanced data with various degrees of skew and has a wide range of application scenarios;Finally,Experiments show that compared with the traditional classifier,the Skip-GANormaly model based on adversarial generation network has more superior classification performance.