On the Network Course Evaluation by Using Nearest Neighbor- Clustering RBFNN and UDM

The network course evaluation indicator system are established in the paper. The large number of representative uniformly distributed samples are designed for training the nearest neighbor- clustering RBF neural network (RBFNN) and solving the problem of RBFNN model´s poor generalization ability. The experiments show the result of nearest neighbor- clustering RBFNN evaluation is very close to the expected result of the experts fuzzy comprehensive evaluation(EFE). The evaluation method realizes the self-adaptive and non-linear approaching ability, meantime conquers the capability limitation of traditional BP neural network and non-preciseness of lacking experiment design, and avoids the subjectivity and uncertainty of traditional evaluation.