Research On Credit Card Fraud Detection Based On Ensemble Learning

With the rapid development of credit card business and Internet technology,the data and information involved in credit card business are complicated,and the payment channels are diversified.Large-scale trend,financial institutions face huge economic losses every year.In addition to the relevant laws promulgated by the national government to protect the interests of financial institutions,it is also essential for financial institutions to establish internal large-scale database management systems,and credit card fraud detection technology has become the top priority in building management systems.Better safeguard the rights and interests of the enterprise itself.Based on the Kaggle platform's latest publicly released credit card fraud detection data set in 2018,this article first normalizes the data set,uses the SMOTE algorithm to perform imbalance processing,features-based probability distribution map(PDF),and based on random forest.The important features are sorted and filtered.Then,the traditional model(logistic regression,random forest),Boosting model(Ada Boost,XGBoost,Light GBM)and LR-RF-XGBoost integrated voting model are used to train the data set.After adjusting and optimizing the parameters,the prediction results of the six types of models are compared and evaluated according to the evaluation indicators(accuracy,recall,F1 value,AUC value,and ROC curve).The example analysis results show that the combined evaluation model of credit card fraud detection based on ensemble learning(LR-RF-XGBoost voting model)has greatly improved the accuracy of credit card transaction detection,well solved overfitting,and predicted results.The problem of instability also greatly improves the generalization ability of the classifier.