Research On Credit Evaluation Of Small And Micro Enterprises

How to use the big data to accurately assess the level of enterprise credit,establish credit evaluation model for financial institutions has a very important significance.Relative to the large and medium-sized enterprises,Small and micro businesses financial information is not perfect,and no large and medium-sized enterprises have obvious assets and brand endorsement,the Small and micro businesses credit is more difficult,therefore,Small and micro businesses loans to financial institutions is always difficult.However,small and micro enterprises have an indispensable position and role in the real economy.How to improve the financing environment of small and micro enterprises,establish accurate credit evaluation model of small and micro enterprises,not only for financial institutions,but also for small and micro enterprises are very important.The commonly used machine learning model,although able to establish the credit evaluation model of credit,and identify whether or not the result,but can not explain every relevant variable for credit and whether it is positive or negative role.On the basis of the commonly used random forest model,this paper combines the nonparametric variable selection method,and further explains the classification results of machine learning through the semi parametric Logistic regression model.The empirical results show that the combination of nonparametric variable selection and random forest model can improve the performance of the ordinary machine learning model,and can enhance the interpretability of machine learning model.Using local constant least squares(LLLS)method for feature selection improves the average accuracy of the classifier more than 1.6%,and the classification performance of random forest is significantly better than other classification models.According to the analysis of influence degree of each index in the process of training for general random forest classification results,and solving the non parametric Logistic regression model of linear correlation coefficient of the variables,the amount of tax payable,the amount of credit and credit Small and micro businesses showed a linear positive correlation,and the correlation is relatively large;consumption,asset liability ratio and Small and micro businesses credit the level of a linear negative correlation,and the correlation between the rate of assets and liabilities for credit level is greater than consumption;nonlinear variables,social pension units should pay the balance with sales revenue for Small and micro businesses credit level related degree.