Application And Comparison Of Machine Learning Algorithms In Credit Card Fraud Identification

In recent years,due to the popularity of credit cards and the rapid development of electronic services,including e-commerce,e-finance and mobile payment,the volume of credit card transactions has increased dramatically,and the risk can not be underestimated.The large-scale use of credit cards and various transaction scenarios,without strict inspection and supervision,will inevitably lead to credit card fraud and billions of dollars of losses and credit risk exposure every year.Therefore,the study of credit card default and anti fraud is of great significance to avoid the losses of financial institutions and individuals,increase the profits of the bank card industry,promote the virtuous cycle of the credit card industry,and maintain the healthy and stable development of the national economy.The core premise of identifying credit card default fraud is to complete the credit risk management of credit card,and the most important thing is to choose the optimal credit risk assessment method.There are many methods of credit risk assessment.At present,the theoretical methods can be roughly divided into "traditional" and"modern".The traditional regression analysis model is simple,convenient and easy to operate,but it requires independent variables,which limits the practical application scenarios.In recent years,scholars have studied a variety of machine learning methods,how to effectively prevent credit risk,improve credit risk assessment methods,and build the optimal anti fraud algorithm model is an important academic research topic.In view of this,this paper applies the algorithm model based on logistic regression,decision tree,random forest,adaptive lifting algorithm,gradient descent decision tree and other statistical methods and machine learning methods for the credit card application acceptance and approval,establishes the personal credit evaluation model for the open source credit card data set,and continuously optimizes the basic model,and then tests with the test set data.At the same time,in order to improve the accuracy of the base model,we try to combine different ensemble learning algorithms to sort the feature importance of the feature variables on the basis of various algorithms,and screen the most important variables that contribute to the binary classification problem.The results show that for the credit card data set used in this paper,random forest classifier and gradient lifting decision tree are superior to other models in accuracy,recall,F1 value,AUC and other indicators.The AUC value of the composite model formed by bagging and boosting is 20%higher than the maximum range of the base model,which shows that the combined model has higher prediction accuracy and better performance.In particular,on the basis of bagging algorithm to complete feature engineering,using gradient lifting decision tree can surpass the combination model of basic model and other integrated algorithms by at least 2‰ in coverage,accuracy,F1 value,AUC value and other indicators,which further shows the significant effectiveness of this combination to improve the classifier effect,It can make the training speed of the model faster under the condition of improving the accuracy.When banks and other financial institutions identify potential default customers,they can fully consider the introduction of such a composite model,which can have a better warning function for credit card fraud default risk,reduce the fraud risk in the early stage,and reduce the supervision cost of managers.