Researching On The Identification Of Fraudulent Accounts In Electronic Banking Based On Federal Learning

At present,people use e-banking more and more frequently and more widely in their work and life,and the number of e-banking fraud crimes is also rising,which brings great destructiveness to people’s life and financial order.In particular,with the integration of social tools and finance,which have become an integral part of popular life,e-banking financial fraud is becoming increasingly prominent in the cross-industry cross-institutional form.How to quickly identify fraudulent accounts and stop criminal activity before it is committed is the key to preventing such crimes before they are committed,and there is a growing call for innovative data sharing and cooperation mechanisms from a technical perspective.Subjecting to data protection laws and regulations,this study proposes an electronic banking fraud account identification method based on the federal learning framework,which tries to realize the identification of fraudulent accounts by training machine learning model by not exchanging private data between the parties and passing only encrypted intermediate parameters.First,building a platform for participants to exchange encrypted parameters,then,using two algorithms,logical regression(LR)and Extreme Gradient Boosting(XGBoost),to build two types of models locally,then,training established models with financial data and financial-social data,and finally,evaluating the recognition of two different data sample patterns with the filtered five indicators.Experimental results show that the recognition effect of using financial-social data(i.e.,after the expansion of sample dimension)is better than that of using only local financial data,which can take into account the dual requirements of protecting data privacy and identifying fraudulent accounts.