Bankruptcy prediction and neural networks: the contribution of variable selection methods

Of the methods used to build bankruptcy prediction models in the last twenty years, neural networks are among the most challenging. Despite the characteristics of neural networks, most of the research done until now has not taken them into consideration for building financial failure models, nor for selecting the variables to be included in the models. The aim of our research is to establish that to improve the prediction accuracy of the models, variable selection techniques developed specifically for neural networks may well offer a useful alternative to conventional methods.