Unsupervised neural networks approach for understanding fraudulent financial reporting

Abstract

PurposeCreditor reliance on accounting‐based numbers as a persistent and traditional standard to assess a firm's financial soundness and viability suggests that the integrity of financial statements is essential to credit decisions. The purpose of this paper is to provide an approach to explore fraudulent financial reporting (FFR) via growing hierarchical self‐organizing map (GHSOM), an unsupervised neural network tool, to help capital providers evaluate the integrity of financial statements, and to facilitate analysis further to reach prudent credit decisions.Design/methodology/approachThis paper develops a two‐stage approach: a classification stage that well trains the GHSOM to cluster the sample into subgroups with hierarchical relationship, and a pattern‐disclosure stage that uncovers patterns of the common FFR techniques and relevant risk indicators of each subgroup.FindingsAn application is conducted and its results show that the proposed two‐stage approach can help capital providers evaluate the reliability of financial statements and accounting numbers‐based decisions.Practical implicationsFollowing the SOM theories, it seems that common FFR techniques and relevant risk indicators extracted from the GHSOM clustering result are applicable to all samples clustered in the same leaf node (subgroup). This principle and any pre‐warning signal derived from the identified indicators can be applied to assessing the reliability of financial statements and forming a basis for further analysis in order to reach prudent decisions. The limitation of this paper is the subjective parameter setting of GHSOM.Originality/valueThis is the first application of GHSOM to financial data and demonstrates an alternative way to help capital providers such as lenders to evaluate the integrity of financial statements, a basis for further analysis to reach prudent decisions. The proposed approach could be applied to other scenarios that rely on accounting numbers as a basis for decisions.

Elucidating the role of topological pattern discovery and support vector machine in generating predictive models for Indian summer monsoon rainfall

“…The GHSOM is a dynamic, multilayer hierarchical and non-fixed structure developed to overcome the major limitations of classical SOM (Rauber et al 2002;Kohonen, 2001) and can generate more detailed clustering than SOMs. Because of their flexibility, GHSOMs have been implemented in many domains, like web mining (Soriano-Asensi et al 2008), data mining (Huang et al 2012;Rauber et al 2002), flow pattern recognition (Tsui and Wu, 2012), forecasting (Guo et al, 2011;Hsu et al 2009;Lu and Wang 2010) etc.…”

Fraudedetectie door tekstanalyse van jaarverslagen

“…Beneish (1997) gebruikt daarnaast het percentage aandelen van het management als variabele voor het detecteren van manipulatie van de Generally Accepted Accounting Principles (GAAP). Verschillende onderzoekers ontwikkelden modellen met financiële ratio's, die een verhouding tussen twee waardes uit de jaarrekening weergeven, voor het detecteren van indicaties van fraude in jaarverslagen (Hoogs et al 2007;Huang et al 2012;Kaminski et al 2004;Perols 2011;Ravisankar et al 2011). Grove and Basilisco (2008) tonen aan dat de financiële informatie alleen niet voldoende is om indicaties van fraude op te sporen.…”

Text mining to detect indications of fraud in annual reports worldwide

“…Ravisankar et al (2011) were able to detect fraud in the financial statements of Chinese companies by applying several machine learning techniques with financial ratio's as input. Huang et al (2012) developed a neural network approach using financial statement variables to aid credit providers in assessing the reliability of Taiwanese financial statements. Their model provides the risk indicators that need further investigation.…”

“…The newest and popular stateof-the-art machine learning technique is deep learning. Deep learning models use neural networks containing more than one hidden layer, opposed to the neural networks popular for experimentation with financial information to detect fraud (Green and Choi, 1997;Fanning and Cogger, 1998;Bhattacharya et al, 2011;Huang et al, 2012). The deep learning models are successfully used for various types of text analysis research tasks.…”

“…In the subsequent years, the financial ratios formed the input for machine learning techniques, such as Decision Trees, Neural Networks and Bayesian Belief Networks, K-nearest neighbors and Support Vector Machines (SVM) (Kotsiantis et al, 2006;Kirkos et al, 2007). In particular, neural networks were a popular machine learning method to detect fraud based on financial information (Green and Choi, 1997;Fanning and Cogger, 1998;Bhattacharya et al, 2011;Huang et al, 2012).…”