Research On The Application Of XGBoost Based Federated Learning In Big Data Risk Control

It is difficult to solve the financing problem caused by information asymmetry only by relying on the traditional financial model,but risk control with the help of big data technology is an effective way to solve such problems.However,as the problem of data security and privacy protection becomes more and more prominent,the problem of data island becomes a stumbling block on the road of fintech development.In order to solve the problems of data security,privacy protection and data isolation,and realize multi-party data sharing to build a better intelligent risk control model,Federated Learning provides a solution.Compared with traditional machine learning models,Federated Learning can realize the establishment of high-performance machine learning models based on multi-party data under the premise of privacy protection and data not going out.This paper takes personal credit loan data of Lending Club platform as an example to establish XGBoost models under Federated Learning and traditional machine learning modes respectively,and evaluates and compares the effects of the models by using AUC,KS and other indicators.The results show that the effect of XGBoost model based on Federated Learning is almost the same as that of XGBoost model based on centralized training.Modeling with data from a single participant is less effective than modeling with data from two participants.This means that Federated Learning can not only meet the dual requirements of data privacy protection and data sharing,but also realize the establishment of a shared model that directly centralizes the data under the ideal performance effect.Therefore,Federated Learning technology has a good development prospect and huge application space in the field of big data risk control.