The robustness test results of the model.

To enhance the accuracy and response speed of the risk early warning system, this study develops a novel early warning system that combines the Fuzzy C-Means (FCM) clustering algorithm and the Random Forest (RF) model. Firstly, based on operational risk theory, market risk, research and development risk, financial risk, and human resource risk are selected as the primary indicators for enterprise risk assessment. Secondly, the Criteria Importance Through Intercriteria Correlation (CRITIC) weight method is employed to determine the importance of these risk indicators, thereby enhancing the model’s prediction ability and stability. Following this, the FCM clustering algorithm is utilized for pre-processing sample data to improve the efficiency and accuracy of data classification. Finally, an improved RF model is constructed by optimizing the parameters of the RF algorithm. The data selected is mainly from RESSET/DB, covering the issuance, trading, and rating data of fixed-income products such as bonds, government bonds, and corporate bonds, and provides basic information, net value, position, and performance data of funds. The experimental results show that the model achieves an F1 score of 87.26%, an accuracy of 87.95%, an Area under the Curve (AUC) of 91.20%, a precision of 89.29%, and a recall of 87.48%. They are respectively 6.45%, 4.45%, 5.09%, 4.81%, and 3.83% higher than the traditional RF model. In this study, an improved RF model based on FCM clustering is successfully constructed, and the accuracy of risk early warning models and their ability to handle complex data are significantly improved.为提升风险预警系统的准确率与响应速度，本研究构建了一种融合模糊C均值（Fuzzy C-Means, FCM）聚类算法与随机森林（Random Forest, RF）模型的新型风险预警系统。首先，基于操作风险理论，选取市场风险、研发风险、财务风险及人力资源风险作为企业风险评估的核心指标。其次，采用准则间关联权重法（Criteria Importance Through Intercriteria Correlation, CRITIC）测算各风险指标的重要程度，进而提升模型的预测能力与稳定性。随后，利用FCM聚类算法对样本数据进行预处理，以提升数据分类的效率与准确率。最后，通过优化随机森林算法的参数构建改进型RF模型。本次研究所选用的数据主要来自RESSET/DB数据库，涵盖债券、国债、公司债等固定收益类产品的发行、交易与评级数据，同时提供基金的基础信息、净值、持仓及业绩数据。实验结果表明，该模型的F1值达87.26%，准确率为87.95%，曲线下面积（Area under the Curve, AUC）为91.20%，精确率为89.29%，召回率为87.48%；相较于传统RF模型，上述指标分别提升了6.45%、4.45%、5.09%、4.81%与3.83%。本研究成功构建了基于FCM聚类的改进型RF风险预警模型，显著提升了风险预警模型的准确率及其处理复杂数据的能力。