论文部分内容阅读
为了提高神经网络分类器的性能,提出一种基于阴影集的训练样本数据选择方法.在阴影集的基础上提出核数据和边界数据的概念.首先通过模糊C均值聚类(FCM)获取样本数据的最优模糊矩阵;然后诱导出相应的阴影集;样本数据结合阴影集构造核数据和边界数据;最后在核数据和边界数据中进行数据选择.利用该方法,结合Iris数据集分别对BP网络、LVQ网络和可拓神经网络(ENN)等分类器进行实验研究.结果表明:该方法能够保留典型的样本,减少训练样本数据的数量;利用该方法所选择的数据对神经网络分类器进行训练,保证了分类器的泛化能力,节约了训练时间,有效提高分类器的性能.
In order to improve the performance of neural network classifier, a method based on shadow set is proposed to select training data samples. Based on shadow set, the concept of kernel data and boundary data is proposed. First, the sample data is obtained by fuzzy C-means clustering (FCM) , Then the corresponding shadow set is induced, and the corresponding shadow sets are induced. The sample data and the shadow set are used to construct the nuclear data and the boundary data. Finally, the data are selected in the nuclear data and the boundary data. Using this method and the Iris data set, , LVQ network and ENN, etc. The experimental results show that this method can preserve the typical samples and reduce the number of training samples, and use the data selected by this method to train the neural network classifier , To ensure the generalization ability of the classifier, save the training time and effectively improve the performance of the classifier.