Cancer Classification by Correntropy-Based Sparse Compact Incremental Learning Machine

Abstract

Cancer prediction is of great importance and significance and it is crucial to provide researchers and scientists with novel, accurate and robust computational tools for this issue. Recent technologies such as Microarray and Next Generation Sequencing have paved the way for computational methods and techniques to play critical roles in this regard. Many important problems in cell biology require the dense nonlinear interactions between functional modules to be considered. The importance of computer simulation in understanding cellular processes is now widely accepted, and a variety of simulation algorithms useful for studying certain subsystems have been designed. In this article, a Sparse Compact Incremental Learning Machine (SCILM) is proposed for cancer classification problem on microarray gene expression data which take advantage of Correntropy cost that makes it robust against diverse noises and outliers. Moreover, since SCILM uses l1-norm of the weights, it has sparseness which can be applied for gene selection purposes as well. Finally, due to compact structure, the proposed method is capable of performing classification tasks in all of the cases with only one neuron in its hidden layer. The experimental analysis is performed on 26 well known microarray datasets regarding diverse kinds of cancers and the results show that the proposed method not only achieved significantly high accuracy but also because of its sparseness, final connectivity weights determined the value and effectivity of each gene regarding the corresponding cancer.