Research on bankruptcy risk prediction of listed companies based on logistic regression model

This paper takes 6,819 listed companies in Taiwan in 2015-2024 as a sample to build a bankruptcy prediction model including 95 financial features. For the first time, SMOTE oversampling technology was combined with logistic regression model, and we found that logistic regression performed well in bankruptcy identification (99.2% accuracy, AUC = 0.983), and achieved 100% recall rate of bankruptcy. Characteristic importance analysis revealed the ratio of net profit to total assets (0.0494), earnings per share for the continuous four seasons (0.0481), and the ratio of total debt to net worth (0.0411) as the core early warning indicators. The study verifies the advantages of the linear model in the linear division scenario of financial data, and provides a methodological reference for the risk assessment in emerging markets.