Research On Online Loan Default Prediction Method Based On LightGBM And BP Neural Network

With the vigorous development of the internet industry,the traditional financial industry has begun to rapidly transform towards internet finance.Online lending has grown rapidly due to its low threshold,fast and convenient operation,and high returns.However,it has lowered the threshold for borrowing and led to a series of illegal collection issues,which have had a significant social impact.How to control default risk has been an urgent problem to be solved in recent years.Due to the diverse forms of online loan data structures,high degree of feature redundancy,and high-dimensional nonlinearity of features,it is difficult to effectively predict defaulters.This article proposes the LightGBM-BP-LDP(Online Loan Default Prediction Method Based on LightGBM And BP)model to address the above issues.The main work is as follows.Firstly,in response to the diverse forms and complex manifestations of online loan data structures,this article conducts missing value processing,data normalization,and feature encoding for data normalization processing;To address the issue of high data redundancy,Pearson correlation coefficient is used to filter the data to reduce redundancy.On this basis,lambda functions are used for feature transformation to provide a data analysis basis for model establishment.Secondly,research on the optimization method of default characteristics for online loans based on LightGBM.Aiming at the problem of low regularization degree of online loan data and low efficiency of feature splitting,data constraint processing is conducted through feature discretization,and the splitting method of Leaf wise is used to improve the splitting efficiency,achieving the effect of enhancing data regularization and optimizing feature selection.Once again,this article constructs the LightGBM-BP-LDP model based on ensemble learning methods.By utilizing the ability of BP neural network to handle nonlinear problems,a LightGBM-BP-LDP model was constructed,and based on ensemble learning methods,online loan default prediction was achieved.Finally,the Lending Club dataset was used for online loan default prediction research,and the effectiveness and accuracy of the online loan default prediction model that integrates LightGBM and BP neural network were verified through comparison.