Improving the Performance of Principal Components for Classification of Gene Expression Data Through Feature Selection

The gene expression data is characterized by its considerably great amount of features in comparison to the number of observations. The direct use of traditional statistics techniques of supervised classification can give poor results in gene expression data. Therefore, dimension reduction is recommendable prior to the application of a classifier. In this work, we propose a method that combines two types of dimension reduction techniques: feature selection and feature extraction. First, one of the following feature selection procedures: a univariate ranking based on the Kruskal-Wallis statistic test, the Relief, and recursive feature elimination (RFE) is applied on the dataset. After that, principal components are formed with the selected features. Experiments carried out on eight gene expression datasets using three classifiers: logistic regression, k-nn and rpart, gave good results for the proposed method.