Research On Credit Rating Of Listed Companies Based On 1D DF-CNN Depth Model

Knowing a company's credit status is very important for banks,investors,and relevant regulatory authorities.However,because China's credit system is very incomplete,the application of credit ratings is limited to the fields of finance,securities,enterprises,governments,and individuals.Credit rating is currently in the early stages of development,and in the previous theoretical research in the field of credit rating,the application of traditional machine learning algorithms and statistical methods in the field of credit rating has been extensively studied.This paper studies how to establish a more mature and practical credit rating model for listed companies.The research plan selected in this paper is to study the importance,feasibility and practicality of the deep learning model-convolutional neural network in the establishment of credit rating models.Since the general convolutional neural network is for two-dimensional image data,the original data set of this paper includes the financial data of 2922 Chinese listed companies in the quarterly financial data 2018-09-30,and the data of each company is presented in one-dimensional data form,this paper proposes a credit rating model for listed companies that uses a one-dimensional convolutional neural network,and selects 38 financial indicators in the company's financial statements as input through correlation analysis.This paper also proposes a new multi-classification method based on the F-score model,and then uses this method to attach a class label to each sample of the dataset used.The F-score model method realized an accurate judgment of the company's credit rating based on its current financial situation.In addition,this paper makes a deeper analysis of the selection of internal parameters such as the size of the convolution kernel and the number of CNN layer groups of the overall network model to determine an optimal network structure for the credit rating dataset.On the basis of this optimal network structure,this paper creates a one-dimensional DF-CNN model by algorithm innovation of one-dimensional convolutional neural network.Through empirical analysis,the classification accuracy of one-dimensional convolutional neural network for credit data is higher than that of traditional machine learning models.Moreover,the classification effect of the one-dimensional DF-CNN model created in this paper is better than the traditional one-dimensional convolutional neural network.