Research on Modeling of E-banking Fraud Account Identification Based on Federated Learning
Boliang Lv, Peizhe Cheng, Cheng Zhang, Hong Ye, Xianzhe Meng, Xiao Wang
Abstract
With the intelligence and groupization of the black market, the characteristics of various fraud methods have become more and more concealed, and cross-industry fraud has gradually become the norm. A single fraud runs through multiple links in social media and banks, and it is difficult for each institution to deal with it based on its own data. Since anti-fraud information involves a large amount of private data, considering that companies are paying more and more attention to the protection of data assets, and the global legal system for privacy data supervision is constantly improving, it is difficult to carry out anti-fraud collaboration among companies. On the premise of satisfying monitoring privacy protection, this paper uses vertical federated learning technology to combine financial and social characteristics to construct a federated learning model of social financial fraud accounts. Compared with the logistic regression (LR) model and the extreme gradient boosting decision tree (XGBoost) model that only use financial feature training, the federated learning model has an accuracy rate of 5.9%, a precision increase of 4.9%, and a recall rate of 5.5% compared with the local model, which verifies the effectiveness of the federated learning technology in improving the model effect.