基于增量学习的混合RBF-ELM网络医学数据分类探析

Research on medical data classification based on incremental learning in hybrid RBF-ELM Network

为提高传统学习算法的数据分类性能,提出了一种基于增量学习的混合RBF-ELM网络(IHRBF-ELM),并应用于医学数据分类.在网络结构的构建上,将RBF隐藏层与ELM隐藏层相级联,即在连接输入层与ELM隐藏层之间加入RBF映射层;在学习算法的实现上,先采用基于势函数聚类的增量学习算法,实现RBF隐藏层高斯核个数及核参数的自动优化估计,并在RBF核空间使用极限学习机优化算法,实现网络输出权值的优化.在不同的医学分类数据集上,通过对混合RBF-ELM网络算法与PFRBF、ELM、HRBF-BP算法进行对比,发现该算法的网络分类精度显著优于其他算法.研究结果对提高传统学习算法的数据分类能力具有参考价值.

To improve the data classification performance of traditional learning algorithms,a hybrid RBF-ELM network(IHRBF-ELM)based on incremental learning is proposed and applied to medical data classification problcems.In the implementation of network structure,the RBF hidden layer is cascaded with the ELM,which means an RBF mapping layer is added between the connecting input layer and the ELM hidden layer;in the implementation of learning algorithms,the incremental learning algorithm based on potential function clustering is first used to automatically optimize and estimate the number of Gaussian kernels and kernel parameters in the RBF hidden layer,and then the extreme learning machine optimization algorithm is utilized to optimize the network output weights.The algorithm was experimentally compared with PFRBF,ELM,HRBF-BP algorithms on different medical classification datasets,and the results showed that the network classification accuracy of IHRBF-ELM algorithm is higher.Thus,the method proposed has good reference value for improving the data classification ability of traditional learning algorithms.