Abstract

Financial accountants falsify financial statements by means of financial techniques such as financial practices and financial standards, and when compared with conventional financial data, it is found that the falsified financial data often lack correlation or even contradict each other in terms of financial data indicators. At the same time, there are also inherent differences in reporting patterns from conventional financial data, but these differences are difficult to test manually. In this paper, the fuzzy C-means clustering method is used to amplify these differences and thus identify false financial statements. In the proposed algorithm, firstly, the normalization constraint of the FCM clustering algorithm on the sum of affiliation of individual samples is relaxed to the constraint on the sum of affiliation of all samples, thus reducing the sensitivity of the algorithm to noise and isolated points; secondly, a new affiliation correction method is proposed to address the problem that the difference in affiliation is too large after the relaxation of the constraint. In the discussion of this paper, most of the information comes from the annual reports of companies, administrative penalty decisions of the Securities Regulatory Commission, and some information comes from research reports made by securities institutions, which are limited sources of information. The proposed method can correct the affiliation to a reasonable range, effectively avoiding the problem that some samples have too much affiliation and become a class of their own and also avoiding the problem that it is difficult to choose the termination threshold of the algorithm iteration due to too little affiliation, and can ensure that the constraint on the sum of affiliation of all samples is always satisfied during the iteration of the algorithm. The method has the characteristics of high recognition accuracy and has the significance of theoretical method innovation.