The Design And Application Of Early Warning Model Of Enterprise Loan Default Risk Based On Stacking Fusion Algorithm

With the increasing downward pressure of China's economy,the financial market environment has gradually deteriorated.The competition among enterprises is becoming more and more fierce,and the cost of external financing is also increasing,which leads to frequent default of enterprise loans in the market.On the one hand,the enterprise loan is an important part of the assets of commercial banks and other financial institutions in China.If the enterprise loan is overdue and the debtor cannot repay the principal and interest,it will seriously damage the profitability and the ability of steady operation of commercial banks and other financial institutions;On the other hand,when commercial banks examine and approve loans,they will increase the interest rate of loans as risk compensation,resulting in the problem of adverse selection,which leads to the increase of financing costs of small and medium-sized enterprises.Therefore,to build an appropriate early warning model of enterprise loan default risk and identify the enterprises with the possibility of loan default is of great help for commercial banks and other financial institutions to optimize the asset health,prevent the occurrence of default risk in advance,and strengthen their own risk control ability.At present,the first mock exam of loan default model in China is mostly based on single model.There are many stacked heterogeneous models lacking in building and forecasting the loan default prediction models.In this paper,python,Excel and other analysis software are used to build the model index system by data preprocessing and feature engineering.The unbalanced data is processed by smote oversampling combined with Tomek link undersampling,and the balanced sample data is obtained.Eight groups of classification models including k-nearest neighbor,support vector machine,naive Bayes,logistic regression,Ada Boost,xgboost,random forest and extratrees were constructed.After searching for the optimal parameters of each model by cross validation method and grid search,the k-nearest neighbor,support vector machine,Ada Boost,xgboost,random forest and extratrees models were selected,Finally,a two-tier enterprise loan default risk early warning model with six groups of models is constructed by stacking fusion of stacking algorithm.The results show that the debt paying ability,profitability,development ability,operation ability,cash flow management ability and risk level of default subject enterprises are significantly worse than those of normal loan subject enterprises;K nearest neighbor,support vector machine,naive Bayes,logistic regression,Ada Boost,xgboost,random forest and extratrees perform better in the balanced data set after comprehensive sampling method;Based on stacking algorithm,the fusion model of knearest neighbor,support vector machine,Ada Boost,xgboost,random forest and extratrees achieves the highest accuracy,recall rate,F1 value and AUC value,which indicates that the two-tier enterprise loan default risk early warning model constructed by stacking fusion of stacking algorithm has excellent classification performance,The multi model stacking fusion strategy of stacking algorithm is feasible.According to the research conclusion,this paper believes that: From the perspective of the model,we should improve the information disclosure and transparency of China's economic market to make the data collection more complete,and multi-dimensional data should be added to the application of the model;From the perspective of the main body of the enterprise,we should constantly strengthen the ability of risk discovery and prevention;From the perspective of commercial banks and other financial institutions,improve the early warning technology of enterprise loan default risk,and use financial technology and artificial intelligence to prevent risks.