ABSTRACT
The emergence of DNA microarray chips has allowed scientists to measure the expression levels of thousands of genes simultaneously. Practitioners have used machine learning techniques to analyze the data from microarray experiments (gene expression data) and make diagnostic and/or prognostic decisions. However, the extremely large number of genes makes traditional machine learning techniques inefficient and ineffective. This chapter shows the importance of taking into account class imbalance when analyzing bioinformatics datasets. It determines whether the order in which feature selection and data sampling are applied is important or not by comparing three approaches developed for classification problems on datasets that exhibit both high dimensionality and class imbalance simultaneously. The chapter utilizes three feature ranking techniques, one form of filter-based subset evaluation, and wrapper subset selection, as well as a commonly used data sampling technique (Random Undersampling).
