A Design Method of RBF Neural Network Based on KNN-DPC

Abstract

In RBF neural networks, the basis functions of hidden layers are often clustered by K-means algorithm. However, due to the K-means algorithm’s dependence on the initial cluster center, it is too sensitive to noisy data. This paper proposes an RBF neural network based on K-nearest neighbors optimized clustering algorithm by fast search and finding the density peaks of a dataset(KNN-DPC). First, the optimized KNN-DPC algorithm is used to cluster data with too many noisy points, then the basis function center of RBF neural network is obtained, finally, the RBF neural network is constructed. The accuracy of this algorithm is verified by simulation experiments, and the results show that the algorithm is effective and practical.