基于免疫聚类的神经网络集成的研究

Research on Neural Networks Ensembles Based on immune Clustering

基于免疫聚类的思想,提出了一种神经网络集成方法。采用轮盘赌选择方法重复地从各免疫聚类中的抽取样本以构成神经网络集成中各个体神经网络的训练样本集,神经网络集成的输出采用相对多数投票法。将基于免疫聚类的神经网络集成应用于中医舌诊诊断,以肝病病证诊断进行仿真。实验结果表明:基于免疫聚类的神经网络集成比基于Bagging算法的神经网络集成能有效地提高其泛化能力。因此,基于免疫聚类的神经网络集成算法的研究是可行的、有效的。

Based on immune clustering, a method for neural network ensemble is proposed. Using roulette wheel selection to sample the data from the immune clustering samples. Then these data are used as training set of individual neural networks in neural net- works ensembles. The output of the ensemble uses majority voting method. Neural networks ensembles algorithm based on immune clustering is applied in TCM tongue inspection diagnose. It uses the hepatic disease symptom as simulation. The experimental result demonstrates that neural networks ensembles algorithm based on immune clustering is better generalization ability than neural networks ensembles algorithm based on Bagging. So, the research on neural networks ensembles algorithm based on immune clustering is possible and effective.