Predicting financial statement fraud using The CRISP approach

 The main purpose of this article is to predict fraudulent financial statements using the CRISP approach. The preliminary data analyzed in this study are from the statistical sample of 164 companies admitted to Tehran Stock Exchange during the period of 2015-2018, which were selected by systematic elimination sampling. The independent variables affecting fraud in this study included 40 financial and non-financial variables that were selected based on antecedent research. Finally, data on variables collected by the library method, based on Crisp approach, to determine the weight and specificity of important variables to the Shannon entropy model and to predict cheating in the top four techniques among intelligence techniques. These techniques include 2 decision trees, neural networks, support vector machines, and the adiabatic hybrid backup vector machine. Using the Shannon entropy out of the 40 research variables, the top 27 variables were identified based on the information profit attribute, which identified the variable cumulative earnings-to-sales ratio as the most important variable in predicting financial statement fraud. After applying the Crisp approach, the results showed that all techniques were capable of detecting financial statements at a relatively high level, and the proposed technique of Adaptive Backup Vector Machine in the training phase with an accuracy rate of 81.69% had higher accuracy and evaluation ability than the other techniques. And this technique correctly identified 82% of fraudulent and non-fraudulent financial statements in the year 2018.