A rough fuzzy neural networks model with application to financial risk early-warning

To overcome the curse of dimensionality, Arough fuzzy neural networks (RFNN) model was proposed in this paper, which combined the rough set theory (RST) and fuzzy neural networks (FNN). First, the models´ input indices (such as financial ratios, qualitative variables et.al.) were reduced with no information loss through rough set approach. And then data based on the reduced indices was employed to develop fuzzy rules and train the fuzzy neural networks (FNN). The new model, which has advantages of both rough set approach and fuzzy neural networks, can not only avoid curse of dimensionality but also prevent “BlackBox” syndrome. The simulation result indicates that the predictive accuracy of the model is much higher. Furthermore, it has characteristics of simple structure, fast convergence speed, and stronger generalization ability etc.