Particle Swarm Feature Selection for Microarray Leukemia Classification

Abstract

In the recent years, DNA microarray has been widely used to investigate genes that cause genetic diseases. Since information from DNA microarray could reveal some interesting relationships between genes and diseases, it has been employed by a number of researchers to classify Acute Lymphoblastic Leukemia (ALL) and Acute Myelogenous Leukemia (AML). As microarray gene expression involves high dimensional features, feature reduction or feature selection is required to ensure efficient classification of ALL and AML. This paper proposes a multi-population particle swarm optimization (MPSO) feature selection approach to identify the most significant subsets of genes for classification of ALL and AML. In this research, MPSO is used to increase the search diversity of conventional particle swarm optimization (PSO). It is combined with the Support Vector Machine (SVM) classifier to form a wrapper feature selection model that can capture the interactions between the classifier and the features. The proposed model is evaluated using 10-fold cross validation. Results showed that MPSO gives a more consistent classification performance than the conventional PSO in ALL and AML classification