基于加权KNN与代价敏感多分支深度神经网络的审计数据异常检测

FRAUD DETECTION OF AUDIT DATA BASED ON WEIGHTED KNN AND COST-SENSITIVE MULTI-BRANCH DEEP NEURAL NETWORK

面对日益提高的审计客观性和不断增长的审计任务,提升审计的效率和质量正成为一种势在必行的趋势。选取电力行业某企业的财务凭证数据为研究对象,针对财务凭证数量多、数据类型多样和数据正负样本比例严重失衡的问题,提出一种基于加权KNN与代价敏感多分支深度神经网络算法。该算法能够有效地缩小核查范围,且得到的支出存在审计疑点的财务凭证中涵盖尽可能多的审计问题,从而更有助于帮助审计人员提高工作效率。通过对比实验,验证了该算法能够有效发现审计疑点及涵盖审计问题,结果优于现有其他方法。

In the face of increasing audit objectivity and increasing audit tasks,it is imperative to improve the efficiency and quality of audit.In this paper,the financial voucher data of an enterprise in the power industry is selected as the research object.Aimed at the problems of large number of financial documents,diverse data types and serious imbalance of positive and negative data samples,a new algorithm based on weighted KNN and cost-sensitive multi-branch deep neural network is proposed.It could effectively reduce the scope of verification,and the obtained financial documents with audit doubtful points covered as much audit problems as possible in order to help auditors improve their work efficiency.Through comparative experiments,it is verified that the proposed algorithm can effectively find audit doubts and cover audit problems,and the results are better than other existing methods.