JIN Dandan,WEN Hui*.Research on medical data classification based on RBF -BP neural network fusion[J].Journal of Yanbian University,2021,47(01):70-74.
基于RBF -BP神经网络融合的医学数据分类研究
- Title:
- Research on medical data classification based on RBF -BP neural network fusion
- 文章编号:
- 1004-4353(2021)01-0070-05
- 分类号:
- TP183
- 文献标志码:
- A
- 摘要:
- 为改善传统的反向传播(BP)神经网络和径向基函数(RBF)神经网络的学习能力和分类性能的不足,提出一种融合RBF网络与BP网络的混合神经网络算法(HRBF -BP),并将其应用到医学数据分类问题中.在网络结构的实现上,将RBF隐藏层与BP隐藏层进行级联融合,即在连接BP网络输入层与隐藏层之间加入RBF核映射层; 在学习算法的实现上,先采用k-均值聚类算法来实现RBF核参数的估计,然后再使用基于随机梯度下降的BP算法实现级联BP网络的权值优化.将该算法与SGBP、KMRB、PFRBF等算法在不同的医学数据集上进行分类实验对比表明,该方法的网络训练精度以及测试精度均优于SGBP、KMRB、PFRBF算法; 因此,该方法对提高BP网络和RBF网络的学习能力和分类性能具有良好的参考价值.
- Abstract:
- To improve the learning ability and classification performance of traditional back -propagation(BP)neural network and radial basis function(RBF)neural network, a neural network algorithm(HRBF -BP)which combines RBF network and BP network is proposed and applied to medical data classification. In the realization of the network structure, the RBF hidden layer and BP hidden layer are cascaded and fused, that is, a new RBF kernel mapping layer is added between the original BP network input layer and hidden layer; in the realization of the learning algorithm, the k -means clustering algorithm is used to realize the estimation of RBF kernel parameters, and then the BP algorithm based on random gradient descent is used to realize the weight estimation of the subsequent cascaded BP network optimization. Compared with SGBP, KMRBF and PFRBF algorithms in different medical data classification experiments, the results show that the network training accuracy and test accuracy of this method are better than SGBP, KMRBF, PFRBF. Thus, this method has a good reference value to improve the learning ability and classification performance of BP network and RBF network.
参考文献/References:
[1] DING S F, SU C Y, YU J Z. An optimizing BP neural network algorithm based on genetic algorithm[J]. Artificial Intelligence Review, 2011,36(2):153-162.
[2] VETELA J E, REIFMAN J. Premature saturation in back -propagation networks: mechanism and necessary conditions[J]. Neural Networks, 1997,10(4):721-735.
[3] BHAYA A, KASZKUREWICZ E. Steepest descent with momentum for quadratic functions is a version of the conjugate gradient method[J]. Neural Networks, 2004,17:65-71.
[4] RIMER M, MARTINEZ T. CB3: An adaptive error function for back propagation training[J]. Neural Processing Letters, 2006,24(1):81-92.
[5] CHEN C H, YAO T K, KUO C M, et al. Evolutionary design of constructive multilayer feedforward neural network[J]. Journal of Vibration and Control, 2013,19(16):2413-2420.
[6] MOODY J, DARKEN C J. Fast learning in networks of locally -tuned processing[J]. Neural Computation, 1989,1(2):281-294.
[7] WU Q, WANG X J, SHEN Q H. Research on dynamic modeling and simulation of axial -flow pumping system based on RBF neural network[J]. Neurocomputing, 2016,186:200-206.
[8] 张爱科,符保龙,李辉.基于改进的模糊聚类RBF网络集成的文本分类方法[J].四川大学学报(自然科学版),2012,49(6):1235-1239.
[9] 韩红桂,乔俊飞,薄迎春.基于信息强度的RBF神经网络结构设计研究[J].自动化学报,2012,38(7):1083-1090.
[10] 黄朝辉,闻辉,车艳.基于势函数聚类的改进RBF网络算法研究[J].延边大学学报(自然科学版),2020,46(2):145-149.
[11] BLAKE C, MERZ C. UCI repository of machine learning databases[EB/OL]. [2020-10-13]. http://archive.ics.uci.edu/ml/35.
备注/Memo
收稿日期: 2020-12-07
*通信作者: 闻辉(1981—),男,博士,副教授,研究方向为机器学习及神经网络.
基金项目: 福建省自然科学基金(2019J01815); 莆田市科技局项目(2018RP4004); 福建省教育科学“十三五”规划项目(FJJKCG20-101)