Classification of microarray datasets using finite impulse response extreme learning machine for cancer diagnosis
Kevin Lee, Zhihong Man, Dianhui Wang, Zhenwei Cao · 2011
Cancer diagnosis is a highly researched field in bioinformatics since the introduction of microarray gene expression technology which allows multiple diseases to be simultaneously tested and compared between healthy and malignant cells. This paper analyses the use of a recently suggested neural network based classifier known as the finite impulse response extreme learning machine (FIR-ELM) for the classification of two binary bioinformatics datasets consisting of microarray gene expressions for leukemia and colon tumor. The FIR-ELM is based on the single hidden layer feedforward neural network (SLFN) whose weights are trained to reduce the effects of noise and improve the robustness of the classifier. The hidden layer of the FIR-ELM is seen to be able to reduce the noise and disturbances from the full microarray dataset which is known to consist of many experimental errors and biases accrued from the production process. Experimental results have shown that the FIR-ELM is capable of achieving good performance compared to conventional classifiers such as the back propagation artificial neural network (BP-ANN), extreme learning machine (ELM), and support vector machine (SVM) which implement a gene selection procedure prior to working on the datasets.