Construction of Credit Rating Models for Listed Companies Based on Machine Learning

In recent years, increased global economic uncertainty and financial market complexity have led to frequent credit risk incidents among listed companies. Traditional credit rating models, which predominantly rely on financial indicators, often prove inadequate for comprehensively assessing enterprise credit risk. This study constructs a machine learning-based credit rating model that integrates financial data, textual information, and network relational data to enhance the accuracy of corporate default prediction. Financial data was sourced from Wind and CSMAR databases, while web crawling techniques were employed to collect news texts. Enterprise relationship networks were constructed using Qichachas data to obtain relevant network metrics. Natural language processing techniques were applied to extract textual features, and graph network analysis methods were utilized to quantify corporate interconnections. Comparative analyses were conducted across multiple algorithms and model performances. The findings demonstrate that multi-source data integration significantly improves predictive accuracy. Graph neural networks exhibit exceptional performance in capturing risk contagion effects, while textual semantic features prove effective in identifying latent risks. This model provides financial institutions and regulatory bodies with a more comprehensive risk monitoring tool, representing a substantial advancement in modern credit rating systems.