基于ICSA优化WELM的不平衡条件下财务风险预测研究

Research on Financial Risk Prediction under Imbalanced Conditions Based on ICSA-Optimized WELM

摘要: 企业财务风险预测对维护市场稳定和投资者权益具有关键作用，但在实际数据中存在类别分布不平衡的挑战，传统预测模型往往对少数类(如ST企业)的识别能力有限。为提高不平衡数据下的预测性能，本文提出一种利用改进乌鸦搜索算法(ICSA)优化加权极限学习机(WELM)的方法。通过动态调整参数和引入莱维飞行机制、多个体加权学习策略和差分进化算法提升算法搜索效率，并结合加权极限学习机对少数类样本进行识别。以2022~2024年A股上市公司为研究对象，采取交叉验证的方法对模型进行性能评估。实验结果表明，本文所提出模型在准确性、召回率、G-mean等关键指标上优于对比模型，能够比较有效地识别ST企业，为财务风险预测提供了更可靠的技术手段。

Abstract: The prediction of corporate financial risk plays a critical role in safeguarding market stability and protecting investor interests. However, real-world datasets often present challenges related to class imbalance, wherein traditional prediction models tend to exhibit limited capability in identifying minority classes, such as Special Treatment (ST) enterprises. To enhance predictive performance under imbalanced data conditions, this paper proposes a method that optimizes the Weighted Extreme Learning Machine (WELM) using an Improved Crow Search Algorithm (ICSA). By dynamically adjusting parameters and incorporating mechanisms such as Lévy flight, a multi-individual weighted learning strategy, and a differential evolution algorithm, the search efficiency of the algorithm is significantly improved. This optimized approach is integrated with WELM to enhance the identification of minority class samples. Using A-share listed companies from 2022 to 2024 as the research sample, cross-validation is employed to evaluate model performance. Experimental results demonstrate that the proposed model outperforms baseline models in key metrics including accuracy, recall, and G-mean, showing effective identification of ST enterprises and providing a more reliable technical solution for financial risk prediction.