Multi-objective feature selection based on quasi-oppositional based Jaya algorithm for microarray data

Abstract

Feature selection is the highly sought-after pre-processing technique for microarray data classification. Microarray data is a phenomenally imbalanced, high-dimensional dataset which requires a specifically designed feature selection technique. A hybrid feature selection (FS) technique is designed in this work for microarray datasets. The proposed FS technique has two phases. In the first phase, aggregate ranking of five well-known filter methods is obtained using the PROMETHEE (Preference Ranking Organization METHod for Enrichment Evaluations) method and is used to remove the unwanted features. The features selected in the first phase are then passed on to the second phase for further refinement. In the second phase, the quasi-oppositional-based multi-objective Jaya algorithm is designed for wrapper-based feature selection. Moreover, a novel quasi-opposite number generation technique for binary numbers is proposed. Ten benchmark microarray datasets are used to evaluate the performance of the proposed technique. Seven other state-of-the-art feature selection techniques are used for comparison and the proposed technique is found to outperform other techniques in terms of classification error rate and the selected number of features. In addition, the proposed FS technique is found to be ten times faster than other FS techniques.