Bank failure prediction with artificial neural networks: A comparative application to turkish banking system In this study, neural network models are introduced and employed for the classification of failed non-failed banks prior to the failure of banks. To perform classification, 36 financial and operational ratios of banks operating in Turkey were used as an input to the models. Two types of neural network structure were employed: Multilayer Perceptron (MLP) and Generalized Feed Forward Networks (GFW). Based on these network structures, ten Artificial Neural Network (ANN) model were constructed having varying number of hidden layers and perceptron in each layer. Rival models are evaluated and compared in terms of classification accuracy. Models are estimated by using the data of 1995-2000 periods, and their out-of-sample forecast comparisons are conducted for the period of 1997-2000 for each year prior to bank failures. In addition to ANN models, to compare the classification performance, discriminant and logistic regression models are employed. Out-of-sample forecast comparison of classification performance of ANN, discriminant and logistic regression models yield that ANN outperforms other models.