基于RFE-LGB算法的上市公司财务造假分析和预测OA

To address the issue of financial fraud prediction in listed companies,a method combining LightGBM and Recursive Feature Elimination(RFE)is adopted for data modeling.LightGBM is known for its low number of hyper parameter,strong robustness,and high sensitivity to imbalanced data.RFE,as an encapsulated feature selection method,can highly match the prediction model used and automatically determine the optimal number of features by setting a feature subset evaluation function as a stopping condition,which has significant advantages in the field of feature selection.In addition,the balanced accuracy(BAcc)is selected as the evaluation index for the predictive performance of the model,and the problem of sample imbalance is solved by adjusting the classification weight parameters of LightGBM.The experimental results on five different industry financial datasets show that the proposed RFE-LGB model exhibits good balance,robustness,and generalization in predicting financial fraud in listed companies.This model can effectively identify key indicators related to financial fraud,and can achieve high prediction accuracy with only a few core features.