基于非平衡数据的机器学习企业财务风险预警模型研究

Research on Machine Learning Models for Enterprise Financial Risk Warning Based on Imbalanced Data

摘要: 制造业企业一般需要较高的研发投入与供应链协同要求，但由于当前国内整体行业的外部环境愈趋复杂、市场竞争激烈，其所面临的财务风险问题亦愈发严峻，因此构建适用于我国制造业企业的财务风险预警模型十分有必要。当前国内制造业的ST与非ST企业数量占比存在极大差距，所以多数研究运用的是非平衡数据。对此本文除了对选取的ST企业进行1:3匹配同等资产规模的非ST企业外，还利用了SMOTE过采样方法对非平衡的样本数据集进行处理。文中共采用了2020年到2023年的168家中国制造业上市企业数据，并综合12种机器学习模型进行了预警效果综合对比。结果表明Extra Trees模型的效果最优，且数据集在经过平衡处理后该模型的预测准确率得到了18%的提升。本研究期望所做的模型研究对于国内制造业企业在财务风险预警防控上具有实际的参考与应用价值，能够助力维护行业与经济的稳健发展。

Abstract: Manufacturing enterprises typically require high R&D investment and supply chain coordination. However, due to the increasingly complex external environment and intense market competition in the domestic industry, the financial risks they face are becoming more severe. Therefore, it is crucial to develop a financial risk warning model suitable for manufacturing enterprises in China. Given the significant disparity in the number of ST and non-ST enterprises in the domestic manufacturing sector, most studies utilize imbalanced data. In response, this paper matches selected ST enterprises with non-ST enterprises of the same asset size in a 1:3 ratio and employs the SMOTE oversampling method to address the imbalanced dataset. The study uses data from 168 Chinese manufacturing listed companies from 2020 to 2023 and compares the performance of 12 machine learning models for risk prediction. The results indicate that the Extra Trees model performs the best, with an 18% improvement in prediction accuracy after balancing the dataset. This research aims to provide practical reference and application value for financial risk warning and prevention in domestic manufacturing enterprises, contributing to the stable development of the industry and economy.