A Pre-investment Enterprise Financial Crisis Identification Scheme Based On KPCA-XGBoost

Enterprise financial crisis is not only a typical two classification problem,combined with early warning mechanism,we can classify it into a multi classification problem.Due to the overlapping of the correlation between the traditional financial index variables,we need a high-precision classification model to predict and identify the financial crisis of enterprises.This scheme selects the specific data of automobile industry from 2010 to 2019,uses KPCA kernel principal component analysis to extract the characteristics of traditional financial indicators,and introduces smote oversampling method to modify the balance of sample data,and finally relies on the xgboost algorithm for modeling.This research shows that: after the data processed by KPCA and smote algorithm is input into xgboost,it has higher recognition ability compared with other mainstream algorithms,the overall accuracy rate reaches 91.25%,and the AUC value reaches 0.87.At the same time,we found that when KPCA is used for feature extraction,we need to pay attention to the advantages and disadvantages of this method.The method of combining KPCA with xgboost is worse than other methods.