DNA Microarray Classification with Compact Single Hidden-Layer FeedForward Neural Networks

Microarrays have been useful in the diagnosis and treatment due to their abilities to survey a large number of genes quickly and to study samples with small amount. With the development of microarray technology, the prospects for effective and reliable disease diagnosis and management can be significantly improved if the classification performance on microarray data is improved. This paper presents an application of the compact single hidden layer feedforward neural networks (C-SLFNs) trained by an improved extreme learning machine (ELM) algorithm to classify microarray data for cancer diagnosis. Experimental results show that the classification accuracy is higher than those achieved by the SLFNs trained by the original ELM and back-propagation (BP) algorithms, and other popular methods for microarray classification such as support vector machine (SVM) and Fisher discriminant analysis (FDA). Moreover, the trained C-SLFNs have a smaller number of hidden units than the SLFNs trained by the original ELM and BP, which results in quick response to new input patterns.