Review on feature selection methods for gene expression data classification

Abstract

Microarray technology makes it easier for scientists to rapidly measure thousands of gene’s expression levels. By analyzing these data, we can find the altered genes, thereby facilitating easy diagnosis and classification of the genetic-related diseases. However, predicting and identifying cancer types is a great challenge in the medical field. Gene expression microarray contains information that can help in this regard, but microarray data have high dimensionality problem which means a large number of genes or features and a small number of samples, also there are redundant and irrelevant features that increase the challenge of microarray analysis. This study reviewed recent studies about methods, algorithms, and limitations of feature selection for microarray gene expression classification. This study compared and focused on four aspects for each related study: datasets, feature selection methods, classifiers, and accuracy results. Feature selection methods are considered as a pre-processing step which plays a vital role in the effectiveness of a classification. This paper showed that applying filter methods such as t-Test, Pearson’s Correlation Coefficient (PCC), and Bhattacharyya distance eliminate irrelevant features that help to increase classification performance and accuracy. Therefore, applying wrapper or embedded methods such as Genetic Algorithm (GA) without applying filter methods in advance could affect the effectiveness of a classification negatively.