Research On Fraud Detection Of Credit Card Based On Rare-Class Classification

In recent years, despite the weak global economic development, China’s credit cardindustry has developed rapidly driven by China’s sustained rapid economy development.Because of the growing popularity of mobile and Internet payment means, thedevelopment of the credit card industry gets continued driving force. The number of creditcards continues to increase. The number of credit card transactions and involving moneyincreases dramatically, which lead credit card risk management become the issues of greatconcern to the industry. As an important part of risk management, credit card frauddetection has significant research value. Due to the low frequency of occurrence andlarge-scale damage, it’s hard to effectively identify credit card fraud using commonmethods. Therefore requires the use of data mining methods for credit card fraud detection.In the research of credit card fraud detection, the main difficulty is that fraud data isrelatively less and it is hard to extract fraud features from normal transaction data. Need totake advantage of the rare class classification method for credit card fraud detection. Thispaper presents the rare-class classification method combining imbalanced data sets processand Adaboost. Process imbalanced data sets by clustering to solve the problem of unevenlydistributed samples, then use Adaboost to classify rare-class. In order to solve theshortcoming of Adaboost that susceptible to the impact of outlier samples, this paperproposes an improving to the weight update mechanism of Adaboost, which can enhancethe weights of category boundary samples and limit the diffusion amplitude of outliersamples’ weights to reduce the influence on the classification accuracy. This paperproposes new construction method of credit card fraud recognition model based onimbalanced data sets processing and classifier training. The experiment results show thatthe model can effectively identify fraud, which provide new ideas for credit card fraudprevention.This paper is divided into six chapters. Chapter1introduces domestic andinternational credit card fraud detection. Chapter2analyzes the rare class classificationmethods, and put forward an idea of the rare class classification. Chapter3studiesimbalanced data set processing method. Chapter4improves the weights update mechanism of Adaboost algorithm. Chapter5proposes a model of credit card fraud detection. Chapter6summarizes the research of this paper and prospects the direction of development.