A hybrid genetic algorithm and support vector machine classifier for feature selection and classification of gene expression

Abstract

Advancement in gene expression technology offers the ability to measure the expression levels of thousand of genes in parallel. Gene expression microarray data is expected to significantly aid in the development of efficient cancer diagnosis and classification platforms. Key issues that need to be addressed under such circumstances are the efficient selection of a small subset of genes that might profoundly contribute to disease identification from the thousand of genes measured on microarrays that are inherently noisy. This research deals with finding a small subset of informative genes from gene expression data which maximizes the classification accuracy. This research proposed a hybrid between Genetic Algorithm and Support Vector Machine classifier for selecting an optimal small subset of informative genes and classifying the optimal subset. Two benchmark data sets were used to evaluate the usefulness of the approach for small and high dimension data. Although, the experimental results showed that the hybrid method performed better than some of the best previous methods on small dimensional data, its performance deteriorated significantly on the higher dimensional data. An improved version of the hybrid method was designed by introducing a new algorithm for features selection based on improved chromosome representation to replace the original algorithm on the hybrid method which appeared to perform poorly on high dimensional data. The results of the gene expression microarray classification demonstrated that the proposed method performed better than the original and the previous methods. The informative genes from the experiment results proved to be biologically plausible when compared with the biological results produced from biologist and computer scientist researches.