批发零售业上市公司财务困境预警--基于RF-VNWOA-LSSVM模型

Financial Distress Warning of Listed Wholesale and Retail Companies: Based on RF-VNWOA-LSSVM Model

本文从国泰安数据库(CSMAR)选取2019—2022年度A股主板被ST或被*ST的33家批发零售业上市公司作为研究样本,选取20个财务指标和9个非财务指标,构建了预警指标体系。为消除非关键特征指标的影响,采用随机森林算法(RF)进行特征值筛选,将筛选的数据集应用于经过优化的LSSVM(最小二乘支持向量机)进行财务预测和预警。实验结果显示,相较于传统的PSO(粒子群优化算法)、GA(遗传算法)以及WOA(鲸鱼优化算法),采用VNWOA优化算法的分类精度分别提高了2.9个百分点、2.9个百分点以及4.35个百分点。综合应用了随机森林和VNWOA优化算法的RF-VNWOA-LSSVM模型在分类精度上相较于RF-费希尔判别法和BP神经网络分别提高了18.75个百分点、8.45个百分点。实验结果表明本文提出的RF-VNWOALSSVM预警模型可以对财务风险进行有效识别。

This study selected 33 A-share main board companies listed from 2019 to 2022 in the Guotai’an CSMAR database,which were either ST or ^(*)ST during this period,focusing on the wholesale and retail industry.Based on the characteristics of listed companies in the wholesale and retail industry and a review of previous literature,20 financial indicators and 9 non-financial indicators were selected to construct an early warning index system.To eliminate the impact of non-key feature indicators,the random forest algorithm was employed for feature selection,and the selected dataset was applied to an optimized LSSVM model for financial prediction and early warning.The experimental results showed that compared to PSO(Particle Swarm Optimization),GA(Genetic Algorithm),and WOA(Whale Optimization Algorithm),the VNWOA optimization algorithm improved prediction accuracy by 2.9%,2.9%,and 4.35%,respectively.The RF-VNWOA-LSSVM model,which combined random forest and VNWOA optimization algorithms,achieved an 18.75%and 8.45%increase in prediction accuracy compared to RF-Fisher discriminant analysis and BP neural network,respectively.The experimental results demonstrate that the proposed RF-VNWOA-LSSVM early warning model can effectively identify financial risks.