摘 要
随着资本市场不断发展，财务造假事件频发，严重损害了投资者利益与市场秩序，如何有效识别和预测财务造假成为监管机构与学术界关注的重点问题。本研究旨在构建一种基于机器学习的财务造假预测模型，通过数据驱动的方法提升预测的准确性与实用性。本文选取2010年至2022年间A股上市公司的财务与非财务数据作为样本，采用逻辑回归、随机森林、梯度提升树（XGBoost）等多种算法进行建模比较，并引入特征选择方法优化模型性能。实验结果表明，XGBoost模型在预测准确率、召回率和F1分数等指标上均优于其他模型，具备较强的判别能力。此外，本文进一步分析了影响财务造假的关键因素，发现公司治理结构、审计意见类型及异常财务指标具有显著预测作用。本研究的创新点在于融合多源异构数据并引入先进的集成学习技术，提升了传统财务舞弊识别方法的效果。研究成果为监管部门提供了有效的技术支持，也为投资者提供了辅助决策工具，具有重要的理论价值与实践意义。
关键词：财务造假预测；机器学习；XGBoost模型；特征选择；公司治理结构
Abstract
With the continuous development of capital markets, financial fraud incidents have occurred frequently, seriously damaging investors' interests and market order. How to effectively identify and predict financial fraud has become a key issue of concern for regulators and academics. This study aims to construct a financial fraud prediction model based on machine learning to enhance the accuracy and practicality of predictions through data-driven approaches. Using financial and non-financial data from A-share listed companies between 2010 and 2022 as samples, this research employs and compares multiple algorithms, including logistic regression, random forest, and gradient boosting trees (XGBoost), while introducing feature selection methods to optimize model performance. Experimental results demonstrate that the XGBoost model outperforms other models in terms of prediction accuracy, recall rate, and F1 score, exhibiting strong discriminative capability. Furthermore, this paper analyzes the key factors influencing financial fraud and finds that corporate governance structure, audit opinion type, and abnormal financial indicators have significant predictive power. The innovation of this study lies in integrating multi-source heterogeneous data and introducing advanced ensemble learning techniques, thereby improving the effectiveness of traditional financial fraud detection methods. The research outcomes provide effective technical support for regulatory authorities and offer auxiliary decision-making tools for investors, holding both theoretical and practical significance.
Keywords: Financial Fraud Prediction; Machine Learning; XGBoost Model; Feature Selection; Corporate Governance Structure
目 录
1绪论 1
1.1研究背景 1
1.2研究意义 1
1.3研究现状 1
1.4本文研究方法 1
2财务造假特征与数据来源分析 2
2.1财务造假的主要表现形式与识别难点 2
2.2数据采集与样本选择标准 2
2.3特征变量的选取与预处理方法 3
2.4数据集的划分与验证策略 3
3机器学习算法在财务造假预测中的适用性分析 4
3.1常用分类算法在财务造假预测中的应用比较 4
3.2模型性能评估指标的选择与设定 4
3.3过拟合与欠拟合问题的应对策略 5
3.4集成学习方法在提升预测精度中的作用 5
4财务造假预测模型的构建与优化 6
4.1模型构建的基本流程设计 6
4.2特征工程对模型性能的影响分析 7
4.3超参数调优方法与实现过程 7
4.4模型稳定性与泛化能力测试 8
结论 8
参考文献 10
致 谢 11
