电商企业财务风险的失衡数据预测研究——基于CSA优化SMOTE与随机森林的融合模型

Study on Predicting E-Commerce Financial Risk from Imbalanced Data—An Integrated Model Based on CSA-Optimized SMOTE and Random Forest

摘要: 鉴于财务风险预测对电子商务行业的发展及电商企业的可持续发展具有极其重要的作用，而现有研究对数据不平衡状态下的预测效果仍存在不足，本文以2020~2024年的电商企业数据为样本开展研究。首先，对比不同采样方式对机器学习分类器效果的影响，确定以SMOTE-RF为基础模型，再引入改进的乌鸦搜索算法(CSA)优化，利用融合模型对数据集进行财务风险预测和分析。实证分析发现，在不平衡数据集上，经乌鸦搜索算法优化后的SMOTE-RF组合分类器整体表现尚可，提高了对少数类的识别效果；在引入改进的CSA后，ICSA-SMOTE-RF模型在保持较高特异度的同时，获得了较高的召回率，对财务风险预测效果有着较大幅度的提升。实证结果表明，本文提出的融合模型能够较好地反映出财务指标之间复杂的非线性关系，为电商企业的风险预测提供了可靠的理论研究方法。

Abstract: In the context of the rapid development of e-commerce industry, financial risk prediction is of great significance to the sustainable development of e-commerce enterprises. However, the existing financial risk prediction research still demonstrates limitations in the prediction effect under the condition of imbalanced data. This study selects e-commerce enterprises from 2020 to 2024 as research subjects. By comparing the effects of different sampling strategies on the performance of machine learning classifiers, SMOTE combined with Random Forest (RF) was determined as the base prediction model. Furthermore, the improved Crow Search Algorithm (CSA) was introduced to optimize model performance, employing a fusion model to predict and analyze financial risks in the dataset. Empirical findings reveal that on imbalanced datasets, the overall performance of the SMOTE-RF combined classifier optimized by the Crow Search Algorithm is acceptable, improving the recognition effect for minority classes. The ICSA-SMOTE-RF model based on improved CSA effectively increases recall while maintaining high specificity, significantly improving financial risk prediction performance. The empirical findings suggest that the proposed hybrid model in this paper effectively captures the complex nonlinear relationships among financial indicators, providing a reliable theoretical and methodological approach for risk prediction in e-commerce enterprises.