基于流形分析与AP算法RBF神经网络分类器

提出了一种基于流形分析与近邻传播（AP）算法的径向基函数（RBF）神经网络分类算法.通过流形分析算法对数据集进行初步处理,然后通过指数函数调整相似度矩阵,再重新进行AP聚类,在此基础上构造RBF神经网络分类器,通过拟合正确率来判断算法是否收敛,并对分类结果运用FMI指标进行评价.实验结果表明：改进算法中RBF网络隐节点数普遍得到增加,使得RBF神经网络拟合精度得以提高;从分类结果可以看出该算法对训练数据集都获得了很好的拟合正确率,对测试数据集也获得了较高的测试正确率.

RBF neural network classifier based on manifold analysis and AP algorithm

A radial basis function(RBF) neural network classification algorithm based on affinity propagation(AP) algorithm and manifold analysis algorithm was proposed.The datasets was preprocessed by manifold analysis algorithm.Then adjusted the similarity matrix by the exponential function and clustered again using AP algorithm.An RBF neural network classifier was constructed based on this,and whether it was converged or not was judged by the correct rate of fitting.The classification result was evaluated by the Fowlkes and Mallows index(FMI) indicators.The experimental results show that the hidden nodes of RBF network are generally increased and the fitting accuracy is improved in improved algorithm.It is seen from the classification results that the improved algorithm obtains a good correct rate of fitting for training datasets and a relatively high correct rate of testing for test datasets.