Development of a Fraud Detection Model Using an Integrated Approach Based on the Factor Analysis Model and the Artificial Neural Network Method in Firms Listed in Tehran Stock Exchange

The main purpose of this research is to propose a fraud detection model using an integrated approach based on the factor analysis model and the artificial neural network method. The network used to predict corporate financial fraud has 17 neurons (selected financial ratios) in the input layer and 1 neuron (corporate fraud status) in the output layer. The conversion function used in the output layer is linear and for the middle layer a non-linear sigmoid function is selected. The neural network used in this research is a feed forward neural network with back propagation algorithm. The statistical population of this study is comprised of the companies listed in Tehran Stock Exchange in the time interval from1392 to 1393. Out of these companies, 140 have been selected as the research sample. The Beneish M-Score model has been used in order to classify the companies with the likelihood of fraudulent and non-fraudulent reporting. According to the Beneish M-Score Model, 78 companies were fraudulent in terms of their reports and 62 were non-fraudulent. For the final selection of the input variables (financial ratios) in the artificial neural network, the confirmatory factor analysis model and the principal component analysis model have been used. The results obtained from the aforementioned models have shown that the reported structure of the neural network model has 7 hidden layer neurons and the momentum learning algorithm has been used for training the network. This algorithm was more precise and functioned better than other reviewed structures. Therefore, it was selected as the final adjustment of the neural network.. The obtained results indicated that the artificial neural network method had a higher performance in this regard; in that the precision of classification of fraudulent and non-fraudulent firms and the overall performance of the artificial neural network method was57and 69%,72,73%,61,62% respectively.