Filter-wrapper combination and embedded feature selection for gene expression data

Abstract

Biomedical and bioinformatics datasets are generally large in terms of their number of features - and include redundant and irrelevant features, which affect the effectiveness and efficiency of classification of these datasets. Several different features selection methods have been utilised in various fields, including bioinformatics, to reduce the number of features. This study utilised Filter-Wrapper combination and embedded (LASSO) feature selection methods on both high and low dimensional datasets before classification was performed. The results illustrate that the combination of filter and wrapper feature selection to create a hybrid form of feature selection provides better performance than using filter only. In addition, LASSO performed better on high dimensional data.