Research On Credit Card Fraud Detection Method Based On Federated Learning

With the development of social economy and financial technology,credit card transactions are more and more widely used in daily life,but the subsequent credit card transaction fraud will cause huge losses to banks or individuals.The existing machine learning methods for fraud detection need to use a large number of data sets as training samples.Because the credit card transaction information is highly confidential data for banks and individuals,different banks are usually not allowed to share their transaction data sets,and a single bank often has a small amount of data,resulting in the isolated data island problem and small sample size problem,In addition,according to the public data,there is a serious imbalance in the credit card transaction data set,and the proportion of fraud samples in the total samples is very low.In view of the above problems,this thesis uses federal learning to solve the problem of joint detection of credit card fraud by multiple institutions.The main research contents include the following two points:(1)Aiming at the small sample size problem in fraud data set,an improved fraud detection model Meta-CNN based on meta-learning is proposed.The model consists of a feature extraction module and a relationship module.The residual network structure is introduced.By improving the triplet loss function,calculates the loss mean of multiple triples,makes the learned features more discriminative for samples from different categories,realizes the effective extraction of sample features under deep neural network,and calculates the feature map and meta-learning query.The similarity of the set samples is classified to provide a model basis for federated learning.Combined with the meta-learning method,the results of each task learning are associated as prior knowledge,so that the model can quickly learn new tasks and solve small sample size problem.(2)Aiming at the isolated data island problem and multi-institutional cooperation divulging user privacy,a cooperative fraud detection model based on Federated learning is proposed,and the differential privacy algorithm is introduced.By improving the system training process and pruning the clients with high contribution,the bank’s data security is further guaranteed and privacy disclosure is prevented.Compared with the traditional machine learning model,the model does not need to train all the data sets of each bank together.It only needs each bank to obtain a global model from the central server,then train its own private data set locally,and then cooperate with each bank to update the global model through the central server,so as to solve the problem of isolated data Island.Finally,the evaluation indexes and privacy performance of the collaborative fraud detection model under different parameter settings are compared and tested.The experimental and analysis results show that compared with the traditional fraud detection methods,the scheme proposed in this paper has significant advantages in classification performance,can better detect credit card fraud transactions,and the security has been improved.