Credit Card Fraud Detection Based On Ensemble Learning

With the development of modern finance and computer science,using credit card has become an important way of consumption.With the development of Internet finance,the popularity of mobile payment and other consumption methods,credit card payment is no longer necessary for cardholders to consume through physical cards.This new way of using credit cards has also brought a new form of fraud.it is now popular to use mathematical statistics models to analyze the transaction amount,code and other data generated by online consumption.By learning the difference between normal consumer behavior and fraud,we can timely predict abnormal behavior and confirm with cardholders to reduce losses.One kind of the fraud prediction model is simple models,such as logistic regression or decision tree,to ensure that the model is interpretable and not easy to overfit;the other is to use complex models such as GBDT or genetic algorithm,but higher precision with easier over-fitting.Focusing on the imbalance of fraudulent behavior data and the limitations of single model,this paper proposes an ensemble model LGBM-RGF-NN using multiple high-precision models to predict fraud.This paper uses 2013 European credit card transaction data,using Light GBM,regularized greedy forest and neural network as the base model.After obtaining the local best parameters of the above algorithm,each model is randomly disturbed based on these parameters to obtain multiple models,and finally integrate all these models.The final experiment compares the performance of the ensemble model of LGBM-RGF-NN with ensemble model of single model and the performance of LGBM-RGF-NN and traditional models,and concludes that the ensemble learning model with multi-model works better.