Research On Credit Card Fraud Detection Model Via On Blockchain And Federated Learning

Credit card fraud seriously affects the normal operation of consumers and financial institutions,causing billions of dollars in losses worldwide every year.Therefore,credit card fraud detection can effectively reduce the losses of financial institutions and cardholders,and ensure the normal operation of financial order.Machine learning algorithms,with their adaptive,efficient,and accurate characteristics,have become the key preferred method for solving this problem compared to deep learning algorithms,which can better handle credit card transaction datasets.However,the credit card transaction dataset has a data imbalance problem,that is,the number of fraudulent transaction samples is much less than that of normal transaction samples,making it difficult to identify fraudulent transactions in a timely manner.In addition,data privacy and security issues limit data sharing between banks and financial institutions,resulting in the current fraud detection system mainly relying on local data of individual banks or financial institutions.In order to solve the above problems,this paper mainly completes the following work:(1)To address the issue of imbalanced credit card transaction datasets,the use of the SMOTE algorithm for preprocessing the dataset has been proposed.This algorithm generates new fraudulent transaction samples while maintaining the distribution of the original dataset,so that fraudulent and normal transaction samples are evenly distributed,with each accounting for 50% of the dataset.This helps to improve the model’s ability to learn the features of fraudulent transactions.(2)To address the issue of data privacy between banks or financial institutions,a credit card fraud detection model called FLRF via on vertical federated learning random forest was proposed,and an improved federated learning random forest model called FLRFB was designed via on the FLRF model.The FLRF model allows banks or financial institutions to benefit from the global model without uploading their local datasets.The FLRFB model adjusts the sample weights and model weights via on the error rate of the samples trained by the FLRF model,further improving the accuracy of the fraud detection model.(3)In order to enhance the security of the FLRF model while ensuring its accuracy for credit card fraud detection,this paper proposes a multi-centralized blockchain-via federated learning random forest framework called BC-FLRF.By replacing the traditional single central server of federated learning with a blockchain network,a multi-centralized credit card fraud detection model is built,reducing the risk of malicious attacks on the central server in centralized federated learning.In addition,the paper proposes a management node verification consensus mechanism called Po M,which elects management nodes on the blockchain to verify local models generated by each bank or financial institution,preventing internal threats to banks or financial institutions and ensuring the accuracy and security of the global FLRF model.