Credit Risk Scoring Model Based On Variable Selection

Credit assessment is important in credit risk management.Aiming at the shortage of current risk assessment methods in bank loan business,this paper proposes innovative methods for event credit data and time-to-event credit data respectively.Based on the event data,we usually face high dimensional credit data with outliers.In this paper,we propose a robust binary classification method,the Lasso regularized smooth 0-1 loss.And the Fabs algorithm is applied to high-dimensional 0-1 loss to efficiently solve variable selection and parameter estimation problems in the highdimensional data and large sample data.The numerical simulation results show that,compared with the performance of Lasso-Logistic regression and Lasso-SVM model,the proposed method that is not easily affected by outliers has good robustness.Applying the proposed method to the analysis of the factors influencing the default of accounts,it is found that the main effect of loan term and the interaction effects between loan term and industry type have an important effect on the default tendency of credit accounts.In addition,the results of reliability evaluation show that,compared with the benchmark method,the Lasso regularized smooth 0-1 loss has the best prediction accuracy under the condition that the number of selected variables is least.Therefore,the analysis results are more reliable and robust.In credit data,the default sample proportion is small and the censored rate is high.Based on the time-to-event data,this paper proposes mixture cure model.The model,which consists of two submodels—binary classification and survival model,can effectively predict not only whether a default will occur but also when this will be.Compared with single binary classification method and single survival analysis model,this model can provide more information.In addition,the MCP penalty is adopted to select important variables.Similarity in signs of two submodels coefficients is promoted to improve the interpretability of the model.The proposed mixture cure model is applied to the default time prediction of credit accounts,and the results show that the proposed method has better prediction accuracy compared with the Lasso-Cox model and the cure model with MCP penalty.The analysis results show that,on the one hand,the default tendency of credit accounts is affected by the main effect of prepayment,loan period,and the interaction effects between prepayment and industry type,between prepayment and loan period,between loan period and loan amount.On the other hand,the main effect of the type of work unit,prepayment,and the interaction effects between prepayment and loan term have an important influence on the time to default.