一种Fibonacci优化理论的改进ELM分类方法

ELM Algorithm Based on Fibonacci Method Optimization Algorithm

作为一种单隐层前馈神经网络,极限学习机(Extreme Learning Machine:ELM)相比传统神经网络算法具有模型简单、泛化能力好、学习速度快等优点,在大规模基因芯片技术的应用中为基因表达数据的肿瘤诊断提供了新的途径,是交叉科学领域新的突破.针对极限学习机随机确定权值,以及其算法存在大量隐层的神经元个数导致算法性能不稳定、分类精度不理想等问题,采用基于优化理论中的Fibonacci序列对ELM隐层节点与偏置进行改进,提出了一种基于Fibonacci优化理论的ELM分类方法(F-ELM).将改进分类方法应用到Hepatitis和Bridges数据集上,实验结果表明,基于Fibonacci优化理论的ELM分类方法性能得到提升,并相对传统的SVM算法、BP和Bayes算法的分类精度较高.

As a single-hidden layer feedforward neural networks, Extreme Learning Machine has the advantages of simple model, good generalization ability, fast learning speed comparing with the traditional neural network algorithm. In large-scale applications of gene chip technology in gene expression data it provides a new way for tumor diagnosis and is a new breakthrough of interdisciplinary field. With the problem of random weights and large numbers of neurons in hidden layer which lead to the unstable performance and unideal classification accuracy in ELM,the paper proposes an improved ELM classification method with the theory of Fibonacci optimization （ F-ELM ）. The algorithm focus on the improvement on the optimization of Fibonacci sequence theory which makes the improvement to the ELM hidden layer nodes and bias. The improved classification method is applied to the Hepatitis and Bridges data set. The ex- perimental results shows that the Fibonacci performance optimization for ELM classification method based on the theory of improved has higher classification accuracy compared to the traditional SVM algorithm, BP algorithm and Bayes algorithm.