Study on regional financial risk early warning system based on uniform design method and Nearest Neighbo-Clustering RBFNN

The regional financial risk early warning indicator system and the normalization of indicators value are established in the paper. By using scientific Uniform Design Method(UDM), a large number of representative uniformly distributed samples are designed for training RBFNN. The experiments show, the difference between RBFNN risk early warning output value and the expert comprehensive fuzzy evaluation (ECF) expectations value has been very small and has achieved the desired targets. The designed early warning method overcomes the difficulties of lacking of large-scale and high quality training samples which causing the RBFNN´s poor generalization ability. And it realizes the non-linear approaching ability. meantime avoids the non-repeatability and instability of traditional experts expectations method