基于数据挖掘的财务欺诈识别

发布时间：2018-05-19 17:15

本文选题：上市公司 + 财务欺诈　；参考：《西南财经大学》2013年硕士论文

【摘要】：上市公司通过财务信息将企业经营状况和经营成果等传递给企业利益相关者,使他们能够了解企业的过去、现在和未来。目前,上市公司财务欺诈问题严重干扰我国证券市场的健康发展,成为证券监管部门、广大投资者等利益相关者关注的焦点问题,建立一套精确的财务欺诈识别模型具有重要的研究意义。财务欺诈的识别方法有很多。本文详细阐述了如下的方法：单变量分析、基于案例的推理、Logstic、线性概率模型、神经网络、多元判别分析、支持向量机模型、决策树、贝叶斯分类模型、主成分回归模型。国内外对财务欺诈的识别大都把所有行业的欺诈公司放在一起研究,而针对按行业分类的研究却很少。本文首先提出研究背景及意义,总结国内外的研究现状,阐明了研究内容和研究方法,然后研究财务欺诈的内涵及界定、财务欺诈的识别方法以及财务欺诈的识别变量。在此基础上,选择了148家财务欺诈企业和与之配对的148家非财务欺诈企业作为研究对象,先用神经网络模型和逻辑回归模型建模,再选出表现最好的模型,最后用这个最好的模型对样本进行分行业建模,分析本研究对财务欺诈识别的效果,并指出研究存在的局限及后续研究方向。研究结果表明,分行业的建模能够明显提高财务欺诈识别模型的识别精度,它可以有效的帮助政府监管部门、投资者和审计部门正确的识别上市公司财务欺诈行为。
[Abstract]:Through the financial information, the listed company passes on the enterprise management status and the management result to the enterprise stakeholders, so that they can understand the past, present and future of the enterprise. At present, the financial fraud of listed companies seriously interferes with the healthy development of China's securities market, and has become the focus of attention of stakeholders, such as securities regulatory authorities, investors and other stakeholders. It is of great significance to establish a set of accurate identification model of financial fraud. There are many ways to identify financial fraud. This paper describes the following methods in detail: univariate analysis, case-based reasoning logstictics, linear probability model, neural network, multivariate discriminant analysis, support vector machine model, decision tree, Bayesian classification model, principal component regression model. The identification of financial fraud at home and abroad mostly studies the fraud companies in all industries, but there are few researches on the classification of financial fraud by industry. This paper first puts forward the research background and significance, summarizes the current research situation at home and abroad, clarifies the research content and research methods, and then studies the connotation and definition of financial fraud, the identification method of financial fraud and the identification variables of financial fraud. On this basis, 148 financial fraud enterprises and 148 matched non-financial fraud enterprises were selected as the research objects. Neural network model and logical regression model were used to model the model, and then the best performance model was selected. Finally, the best model is used to analyze the effect of this study on the identification of financial fraud, and the limitations of the research and the future research direction are pointed out. The results show that the industry modeling can obviously improve the identification accuracy of the financial fraud identification model, it can effectively help government regulators, investors and audit departments to correctly identify the financial fraud behavior of listed companies.
【学位授予单位】：西南财经大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：F275;F276.6;F832.51

【参考文献】