上市公司财务数据异常识别研究的文献综述  

Literature Review on Financial Data Anomaly Detection in Listed Companies

作　　者：郭晓雨 GUO Xiaoyu(China Institute of Atomic Energy,Beijing 100000,China)

机构地区：[1]中国原子能科学研究院,北京100000

出　　处：《信息与电脑》2025年第6期174-176,共3页Information & Computer

摘　　要：综述针对上市公司财务数据异常识别,梳理了从传统统计模型到机器学习模型的发展脉络,总结了各模型特点及应用效果。研究发现,早期识别主要依赖Beneish的M-score等传统统计模型。随着技术的发展,神经网络、支持向量机和集成学习等机器学习方法逐渐成为主流。然而,现有研究在财务指标选取上过于依赖个人经验,在模型构建时未充分考虑行业差异,限制了识别准确率的提升。针对这些问题,文章介绍了融合特征体系构建和二层混合模型等改进方法,并结合当前研究,阐述了数据处理优化、模型构建创新和评分卡应用等方面的进展,展望了基于数据细分和深度学习的未来研究方向,为促进该领域研究提供参考。The review addresses the identification of financial data anomalies of listed companies,combs through the development lineage from traditional statistical models to machine learning models,and summarizes the characteristics and application effects of each model.It is found that early identification mainly relies on traditional statistical models such as Beneish’s M-score.With the development of technology,machine learning methods such as neural networks,support vector machines and integrated learning have gradually become mainstream.However,existing studies rely too much on personal experience in the selection of financial indicators and do not fully consider industry differences in model construction,which limits the improvement of recognition accuracy.To address these issues,the article introduces improvement methods such as fusion feature system construction and two-layer hybrid model,and combines with current research to illustrate the progress in data processing optimization,model construction innovation and scorecard application,and looks forward to the future research direction based on data segmentation and deep learning,which will provide a reference to promote the research in this field.

关 键 词：上市公司 财务数据异常识别 统计模型 机器学习 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]