Research On Enterprise Credit Assessment Based On Machine Learning Methods

With the rapid development of China’s economy,the openness of China’s financial market gradually opened up,the non-performing default rate brought about by venture capital is also increasing year by year,so it is particularly important to establish an effective enterprise credit assessment system and evaluation model.The research of this paper mainly includes four chapters: introduction,machine learning theory,empirical analysis and model conclusion.First of all,according to the principles of accuracy,system,science and operability of credit assessment,a credit evaluation index system containing a total of 59 indicators,such as current ratio,total asset growth rate and asset return rate,is constructed.Using My SQL technology,the sample data of 4082 listed companies operating in 2020 from Cathay Pacific Database,Reith Database and China Economic Network Statistics Database are integrated,cleaned,interstitial before and after two phases,and the missing value analysis of data is carried out,outlier analysis and multi-colinear problem testing.Secondly,the main component method is used to reduce the dimension of the data,eliminate the common linear problem between the indicators,according to the characteristic value close to 1,extract 24 comprehensive indicators,list the linear expression between the main component and the indicator,and calculate its score value,the score value of the sample as the characteristic attribute of the subsequent model.Thirdly,the training set and test set sample are divided according to the default sample and the non-default sample,according to the ratio of 7:3,and then by adjusting the parameters of neural network model and decision tree model,the comparison and analysis is carried out,and two hidden layers are established,the activation function is logical S-shaped,and the initial learning rate is 0.4 BP neural network model.Then the growth law is CHAID,the maximum tree depth is 3,and the maximum number of iterations is 100 decision tree model.Then,multi-model comparison is made by Kneighbor algorithm,simple Bayesian algorithm,support vector machine algorithm,logistic regression model and multi-discrimination analysis.Finally,the model is evaluated by confusing the matrix to calculate the accuracy of each model,the first type of error rate,the second type of error rate and the geometric mean of the sample.The results show that the accuracy of the two-layer BP neural network and decision tree model is 0.982 and 0.984 respectively,the accuracy is higher in many models,and the two types of error rate of decision tree model are 0.0038,0.2187,and the two types of error rate are the lowest,so the decision tree shows the best comprehensive performance in enterprise credit evaluation.Compared with the traditional logistic regression model,machine learning method is better than logical regression model for enterprise credit evaluation in the case of large samples.