Default Prediction of Internet Finance Users Based on Imbalance-XGBoost

Abstract

Fast and accurate identification of financial fraud is a challenge in Internet finance. Based on the characteristics of imbalanced distribution of Internet financial data, this paper integrates machine learning methods and Internet financial data to propose a prediction model for loan defaults, and proves its effectiveness and generalizability through empirical research. In this paper, we introduce a processing method (link processing method) for imbalance data based on the traditional early warning model. In this paper, we conduct experiments using the financial dataset of Lending Club platform and prove that our model is superior to XGBoost, NGBoost, Ada Boost, and GBDT in the prediction of default risk.