The Application of Neural Networks to Predict Fraud: Case Study of Tehran Stock Exchange

Anticipated financial condition and results forecast for fiscal fraud is essential in today's economic world. Forecast such fraud through qualitative and quantitative methods auditors is difficult, that research using neural networks to predict fraud in the Tehran Stock Exchange, which deals with the period 2004–2013 have been active in this market. In this respect, two different types of neural network learning algorithms have been used before, but worthy. The initial study population consisted of 74 pairs of cheaters and non-cheaters now that the restrictions due to the 56 pairs of healthy companies and dishonest achieved and algorithms used in MLP neural network (MLP) and network of the Radial Basis (RBF) is. The first of back-propagation learning algorithm for network training error (BP) and for the second training network algorithm (supervised and unsupervised) is used. The results show the high efficiency of neural network model is provided so that the correct prediction networks used for Multilayer Perceptron network and radial basis function network, respectively 936% and 917% network shows. The Multilayer Perceptron Neural Network Failure to healthy companies and dishonest relatively higher radial basis function network shown so that the forecasts made by the MLP neural network of carefully rather than network-based function is radial.