DE-SVMRank: a differential evolution algorithm with a rank-based feature selection process for microarray data classification

Abstract

DNA Microarrays are powerful tools to analyze and identify certain disease from the expression level of the genes in tissues samples. Many machine learning techniques are suitable for building predictive models to classify microarray samples into different biological categories. The accuracy of the predictive model may benefit from a relevant feature selection method and even more, if the features are ordered in terms of its relevance. In this paper, we propose a rank-based method to create the initial population in a Binary DE-SVM based algorithm used to build a predictive model. The new algorithm (DE-SVMRank) is evaluated in terms of the achieved accuracy by the predictive model and also, the execution time required to complete the maximun number of iterations. Experimental results on public-domain microarrays show that our proposal reduces the computational time in comparison with a similar approach while providing highly competitive results.