Credit Risk Assessment Of Small And Medium-Sized Enterprises Based On Graph Neural Network

As a major participant in economic activities,the credit risk of enterprises has always attracted the attention of academic and financial circles.The proportion of small and medium-sized enterprises in the national economy is becoming more and more important,and they are an important part of my country’s economic development.Therefore,it is particularly important to improve the credit risk assessment capabilities of small and medium-sized enterprises.Traditional evaluation methods rely on the subjective judgment of experts.Faced with such nonlinear,unstable,and volatile time series data,it is often time-consuming and laborious,and cannot quickly give credit ratings for various enterprises.With the rapid development of computer technology,machine learning has shown excellent results in many fields,and some classic algorithms have had an important impact on the financial field.In recent years,graph neural networks have become a popular topic in the field of machine learning,and have been widely used,especially in many economic and financial problems.This paper uses the graph neural network as the technical framework,integrates the internal relationship between corporate financial indicators,and proposes a model for corporate credit risk assessment.Mainly done the following research work:First,based on previous experience,we screened out 29 corporate financial data indicators,abstracted each indicator into a vertex,deeply analyzed the relationship between indicators,and constructed an indicator similarity matrix.On this basis,we used the maximum spanning tree algorithm to realize the Graph structure mapping;secondly,in the representation learning stage of the mapping graph,a graph neural network model is built to obtain its embedded representation.Expand the feature vector of each node to 32 dimensions,perform three Graph SAGE operations on the graph,perform pooling Pool operations on the results obtained respectively,and average the three final output feature vectors to obtain the embedded representation of the graph;finally use A two-layer fully connected network builds a classifier to complete the prediction task.Experimental results on real enterprise data show that the model proposed in this paper can better estimate the multi-level credit level of enterprises.Furthermore,the graph mapping of the tree structure deeply depicts the internal relationship of the company’s various index data,and the classification effect of the model according to the evaluation standard such as ROC is remarkable and has good "robustness".