Small Enterprise Default Discrimination Model Based On Optimal Combination Of Sub-feature Variables

The default discriminant model of small enterprise provides important basis for bank loan decision and enterprise financing.In this paper,the sub-feature variables after the features binning are used to construct the small enterprise default discrimination model.This paper constructs a small enterprise default discrimination model based on the optimal combination of sub-feature variables,including five chapters:the first chapter is the introduction,the second chapter is the basic principle of default discrimination model based on the optimal combination of sub-feature variables,the third chapter is the construction of the model,the fourth chapter is based on the empirical study of Chinese small enterprises,and the fifth chapter is the conclusion.The research focus of this paper includes two aspects:one is how to divide a feature into different sub-feature variables.Any default discrimination model needs variables,and the accuracy of default discrimination model constructed by different variables is different.For example,the feature of"income"can be divided into"high income","medium income"and"low income".Customers with"high income"tend to have stronger solvency and are less likely to default.However,the"revenue"is divided into several intervals.When the critical points of interval division are different,the default identification ability of sub-feature variables is different.The second is the selection of the optimal combination of sub-feature variables.m feature variables,there will be 2~m combination of feature variables,different combination of feature variables to build different default discrimination model,model discrimination accuracy will be different.The innovation of this study are two:First,in terms of the construction of sub-feature variables,the optimal partition point of a feature numerical interval is deduced according to the condition that the gini variance average maximum,which has the maximum default discrimination ability,and thus get different sub-feature variables of the a feature numerical interval.The second is to deduce an optimal sub-feature variables combination based on the minimum weighted error of Type I Error and Type II Error among m(m+1)/2 sub-feature variables combination composed of m sub-feature variables,and establish the corresponding default discrimination model.Research shows that:First,in the optimal combination of sub-feature variables composed of 32 key sub-feature variables,small enterprises are more likely to default when the feature value of"credit card record of legal representative"is the worst record,that is,"with default record",the feature value of"inventory turnover rate"is in the lowest turnover interval[0.00,5.87]and the feature value of"per capita disposable income of urban residents"is in the lowest income range[3058.00,9692.40).Second,the lowest range[3058.00,9692.40)and the lower range[9692.40,10687.56)of"per capita disposable income of urban residents"both have an impact on the default of small enterprises,ranking 3rd and 18th respectively in the order of importance among 32 influencing factors.It indicates that the feature corresponding to these two sub-feature variables"per capita disposable income of urban residents"is an important feature.Similar features include"growth rate of business income"and"monthly household income"of enterprise heads,and they all have two sub-feature variables that have a direct impact on the default status.Third,in the influencing factors of default risk of small enterprises,the importance of non-financial factors accounts for 51.90%,which is more important than internal financial factors and external macro conditions.Fourth,the default discrimination accuracy of this research model is higher than that of DT,KNN,SVM,LR,LDA,DNN and other big data models.