Initializing multilayer neural networks with fuzzy logic

The authors have developed a neuro-fuzzy system that initializes a structured neural network with a fuzzy logic system that is based on expert knowledge. The neural network gains precision through adaptive learning, and is then converted back into a set of fuzzy rules for ease of understanding. The authors discuss a bond rating application that uses this process. The system produces bond ratings that closely match those of human experts, and has higher precision and better generalization than a simple three-layer neural network. The system also makes it easier to understand the neural system's reasoning by translating it into the fuzzy inference format.<<ETX>>